Beruflich Dokumente
Kultur Dokumente
An Introductory Course
Second Edition
C.A. Brebbia & J. Dominguez
VV I I PRESS Boston, Southampton
Computational Mechanics Publications
? A Brebbta
I. Dominguez
Wessex Institute of Technology Escuela Tecnica Superior de Ingenieros Industria
les
Ashurst Lodge University of Seville
Ashurst Av. Reina Mercedes, s/n
Southampton, SO40 7AA, UK
41012, Seville, Spain
Published by
WIT Press/Computational Mechanics Publications
Ashurst Lodge, Ashurst, Southampton, SO40 7AA, UK
Tel: 44@I703 293223; Fax: 44@I703 292853
E-Mail: witpress@witpress.com
http://www.witpress.com
For USA, Canada and Mexico
Computational Mechanics Inc
25 Bridge Street, Billerica, MA 01821, USA
Tel: 978 667 5841; Fax: 978 667 7582
E-Mail: cmina@ix.netcom.com
British Library Cataloguing-in-Publication Data
A Catalogue record for this book is available
from the British Library
ISBN 185312160 6 WTTPress/Computational Mechanics Publications, Southampton
ISBN 1 56252 087 3 Computational Mechanics Inc, Boston
Library of Congress Catalog Card Number 91-77179
No responsibility is assumed by the Publisher, the Editors and Authors for any i
njury and/or
damage to persons or property as a matter of products liability, negligence or o
therwise, or from
any use or operation of any methods, products, instructions or ideas contained i
n the material
herein.
WIT Press/Computational Mechanics Publications 1992
Reprinted 1998
Printed and bound in Great Britain by Print in Black
All rights reserved. No part of this publication may be reproduced, stored in a
retrieval system,
or transmitted in any form or by any means, electronic, mechanical, photocopying
, recording, or
otherwise, without the prior written permission of the Publisher.
to
Alexander, Beatriz, Isabel and Pelayo
DISKETTE
A diskette containing all the source codes in this book is available in two form
ats,
?/? inch or 514 inch, for IBM PC, PS2, or compatible machines.
Please apply to:
Wessex Institute of Technology
Ashurst Lodge
Ashurst
Southampton
SO4 2??
U.K.
Tel: 44@O03 293223
Fax: 44@O03 292853
Please send a cheque or money order for ?28 (or the equivalent in dollars) and
137
220
241
the meshing. In addition there is no need to use elements on the planes of symme
try.
. Figure 3 describes a turbine blade and its base. Notice that discontinuous ele
ments
allow for a simple mesh grading. The reason why these elements are possible in b
ounboundary elements is explained in some of the chapters in this book. From the us
er's point
of view they offer many advantages in terms of alterations of meshes and general
versatility. Figure 4 shows some Von Mises surface stresses concentration at the
root of the blade [2].
Why Boundary Elements?
Figure 1. Analysis of a turbine blade using FliM and BHM. Notice that a
variation in the configuration of cooling elements creates difficulties for the
I- 1?
code (from a colour original)
Figure 2. Cylinder wilh a cylindrical perforation. The boundary clement mesh
represents discretization of one eighth of the problem
Why Boundary Elements? 3
More complex three dimensional structures such as the complete crankshaft
model shown in figure 5 can be discretized relatively easily using a combination
of continuous and discontinuous elements [2]. The model shown in the figure
consists of approximately 2,000 surface boundary elements and each throw of the
shaft has been represented by a boundary clement zone (similar to a finite eleme
nt
substructure or super element) thus making a total of approximately 10,000
degrees of freedom. As only the surface of the shaft has been represented using
elements the modelling time is quite rapid and the element mesh can be autoautomatically created from a boundary model originating in a CAD system. The sol
usolution of problems of this size can nowadays be easily accomplished on the new
generation of powerful engineering workstations. (This model was run on an IBM
RS6000). This example demonstrates that computer time is no longer a primary
concern in boundary element analysis, particularly as it is anticipated further
increases in performance for workstations and all other computers within the nex
t
few years.
The model in figure 5 allows the user to analyse the overall behaviour of the
crankshaft under different loadings. Afterwards part of the crankshaft can be
Figure 3. Discretization of a turbine blade using discontinuous elements (from
a colour original)
Why Boundary Hlemenls?
Figure 4. Von Miscs surface stresses at the root of the blade (from a colour
original)
Figure 5. Complete crankshaft discretized into boundary elements
Why Boundary Elements?
Figure 6. Solid view of the crank throw
Figure 7. Wire frame view of crank showing oil passing through the crank
6
Why Boundary Elements?
studied in more detail to find the effect of oil holes, fillets, etc. on the str
ess
distribution. Figure 6 shows a model for part of a crankshaft discretized with a
finer mesh than the one in figure 5. This crank throw has an oil hole as shown
in the wire-frame diagram in figure 7. The interesting feature of this problem i
s
that small details can be easily represented without causing a massive increase
in
the cost of the analysis. Note that the elements describing the oil hole in figu
re
7
only describe its surface and do not intersect the elements on the exter
ior part
of the model except where the oil hole penetrates the surface.
It is evident from these examples that boundary elements are an ideal tool for e
nginengineering design mainly because it is easy to generate the data required to ru
n a problem
and carry out the modifications needed to achieve an optimum design. With
computer costs declining while engineers' time becomes (or should become!) more
expensive, the saving in engineers' time is of primary importance. (Also, engine
ers
need relief from the dreary task of preparing finite element data.) More importa
nt
still, any tool that can shorten the 'turn around' time taken by the analysis an
d
design can bring forward the completion date of a project.
The future of BEM in engineering is promising and will continue to be so as
long as the developers do not alienate the users by producing codes which arc
unreliable or cumbersome to use. Most of the advantages of BEM arc related to
its more complex mathematical foundations. This provides a high degree of
versatility and accuracy in well-written codes but can have disastrous consequen
ces
in the case of poorly written BEM codes. The BEM is more susceptible to errors
when the appropriate numerical techniques arc not used and it is then important
for developers to understand properly the theory of the method.
Although better computational performance is important in BEM, particularly
for three dimensional problems, improvements in CPU times should not come at
the expense of precision and accuracy. For instance, applying coarse numerical
integration techniques to BEM codes can result in large savings in computer code
s
and give reasonable results in many cases. For other cases however the solution
may be of very poor accuracy or give non-convergent results. This makes such cod
es
unreliable.
Another important advantage of BEM over FEM is when analysing problems
with stress (or flux) concentration. Many such studies have now been
carried out and they tend to demonstrate the high accuracy of boundary elements
for problems such as re-entry corners, stress intensity problems and even fractu
re
mechanics applications. It is not our intention in this introduction to review a
ll
these studies but rather to point out the difference in results that can be obta
ined
using one or the other numerical method. As an illustration the finite
element solutions found along a line in the neighbourhood of a re-entry corner
(figure 8) of a pressure vessel is shown in figure 9. The problem was also analy
sed*
using a photo-clastic model and boundary elements. Results for a finite clement
mesh consisting of approximately 500 degrees of freedom F9 elements) and using
eight nodes elements are compared against BEM solutions obtained using only
20 elements. It is evident from the figures that while the 69 elements finite el
ement
results show lack of equilibrium in the domain as well as on the boundary,
reasonably accurate solutions were obtained using boundary elements. It was only
Why Boundary Elements?
6 in radius
tlllttllttlltlltlltltllllX
Internal pressure
2 61 psi
Line of interest
for stress output
Dimensions in inches
Geometry of the region under consideration
Line of
interest
69 finite elements mesh
(with mid side nodes)
approximately 500 degrees of
freedom
Figure 8. Re-entry corner in pressure vessel
Why Boundary Elements?
20 boundary elements mesh,
120 degrees of freedom
Figure 8 continued
when using a very refined finite element mesh that the FE results were in agreeagreement with the boundary element and photo-elastic model solution i.e for res
ults
obtained using 240 elements (and 1,500 degrees of freedom.) For a full discussio
n
of these results the reader is directed to references [3] to [5].
The development of more powerful hardware specially parallel and vector proprocessing computers favour the use of BEM. These computers are better suited to
deal with the fully populated matrices and the type of operations which are
characteristics of boundary elements. The applications shown in figure 10
demonstrate the use of boundary elements for solving a non-linear problem, i.e.
the contact analysis of a connecting rod. The model in this case was generated o
n
a CAD system and the geometry automatically passed to the BEASY system and
meshed. Notice that only one quarter of the rod needs to be discretized due to
symmetry. The solution of this analysis is shown in figure 11 where contact sursurface gap between the pin and connecting rod is clearly seen.
Problems other than stress or temperature analysis can be solved using
boundary elements. Typical applications include torsion, diffusion, seepage, flu
id
flow and electrostatics. Corrosion engineers have used the method to design bett
er
cathodic protection systems for offshore structures, ships and pipelines. Many o
f
these structures are basically three dimensional and the region of interest exte
nds
to infinity. Consequently they could not be effectively analysed before the deve
lopdevelopment of boundary elements. Early attempts to use finite differences or fi
nite elements
to solve these problems met with little success. For these cases the computer mo
del
has to represent the potential field around the structure, representing the shie
lding
effect of the structural geometry and the effect of the different materials invo
lved.
Unlike a structural model the cathodic protection model is concerned with the
seawater around the structure and the interface between the seawater and the
structure. Hence the use of FEM to analyse the problem would require the
subdivision of the seawater surrounding the structures which is a Herculean task
.
Why Boundary Elements?
S 5 8 8
(isd)ssans
? ?
1
tl
I
I
|
I f 1
1 1 s
41
f IS
5 S 2
? ? ?
u. u. @
HI
.. ,.,,,
i ,
:
1
j
j
d
?
/
if
J
/
I
&
?
H
X ?
IFigure 10. Quarter model of a connecting rod
1-..-.?-1.-7
N'tM6lDSE*07
Figure 11. Final contact configuration after load application showing effective
Figure 12. The CONOCO Hutton TLP (Tension Leg Platform)
12
Why Boundary Elements?
Figure 14. Model of the TLP showing contours of potentials
The use of boundary element method represents the only practical solution for
this problem. The advantage of the method is that only the structure needs to be
defined as the BF.M automatically takes care of the field i.e. the scawater
extending to infinity. Figure 12 shows the First three dimensional BEM cathodic
protection application which was the study of the tension leg platform (TLP) bui
lt
by CONOCO in the Hutton Field in the North Sea. Figure 13 shows the
discretization of a quarter of the structure into the boundary elements used in
the
analysis and figure 14 the results obtained for the potentials on the surface fo
r a
particular configuration of the improved code system used. Since then the bounda
ry
element method has become the key to the successful and practical analysis of
cathodic protection systems and further work has been carried out in this regard
particularly at the Computational Mechanics Institute, Southampton, UK. A
system is now available which allows the corrosion engineer to evaluate design
options, look at problem areas, interpret experimental observations, optimize th
e
design and predict with accuracy and confidence the degree of protection and lif
e
expectancy of a cathodic protection system.
The advances made in cathodic protection modelling using boundary elements
arc just one of the applications of the technique for systems extending to infin
ity.
The method is nowadays extensively used in other problems with infinite or
semi-infinite domains such as those occurring in gcomechanics, ocean engineering
,
foundations, aerodynamics, flow through porous media and many others.
This brief introduction has attempted to point out the advantages of BEM for
a wide variety of engineering problems and the reason why the method should be
taught on an undergraduate as well as a graduate level. University courses shoul
d
include the fundamentals of the method and provide workshops on applications
while short courses with hands-on applications will help to bring the method to
the attention of practising engineers. This book has been written to provide a
Why Boundary Elements? 13
simple and up lo dale introduction to the method to help popularize the techniqu
e
amongst engineers.
The future of BEM hinges on its acceptance by practising engineers, in
particular as a design tool. Developers should aim to make the method
more accessible to engineers by writing codes which are easy to use and by
explaining the fundamentals of the method on the basis of engineering rather tha
n
mathematical concepts. This book has been written in a form that can be used as
a textbook at undergraduate or graduate level and for the engineer in practice
who wants to learn the fundamentals of the technique unaided. Of particular
interest is the way in which the mathematics concepts are introduced and
immediately applied in simple computer codes. These codes D for potential and 2
for elasticity) will facilitate the comprehension of BEM.
This book is based on the authors' many years experience as researchers and
teachers of boundary elements. It is designed to teach in the most effective man
ner
the fundamentals of the method rather than to attempt to demonstrate erudition
on the subject. Many topics have been deliberately omitted to avoid confusing
the reader. The essentials however are all here. It is now left to the reader to
build
on this knowledge.
References
[I] Brebbia, ? A., J. Tclles and L. Wrobel, Boundary Element Techniques - Theory
and
Applications in linaineerina. Springer-Verlag, Berlin and NY, 1984.
[2] Trcvelyan, J. BEASY: Self-Teaching Guide. Computational Mechanics Publicatio
ns,
Southampton and Boston, 1991.
[3] Floyd, ?. ? The Determination of Stress using a Combined Theoretical and Exp
eriExperimental Analysis Approach, Computational Methods and Experimental Measureme
nts,
I'roi. 2nd Int. Conf. June/July I9H4 (C. A. Brebbia, Ed.) Springer-Verlag, Berli
n and
I /j2
N + A2u? \dx2
A.2)
1.1. Fundamental Concepts
15
This operation is culled an 'inner4 product in mathematics and although does not
imply any new concepts, allows us to investigate the properties of the governing
equation. This is done by integrating by parts terms with derivatives in the abo
ve
expression. In this case one can only 'manipulate* in this manner the first term
,
i.e. d2u/dx2, which gives
+tfubydX + \0 A.3)
dx
J
Id J
Notice that Ihc integration by parts has produced two terms, one in the domain
with first derivatives of ? and \\\ and the other on the boundaries (which in th
is
case are simply the two points .v = 0, .v = I).
Furthermore, if the w function has sufficient degree of continuity one can
integrate by parts again to obtain
A.4)
Expression A.4) is of course equivalent lo A.3) but here not only has one
0
passed all derivatives to the newly defined w function but the two terms at x
and .v = 1 give us an insight into Ihe boundary conditions required to solve the
problem. In this case,
? or
needs lo be known at .v = 0 and x =1
A.5)
dx
Notice that the w function which in principle was an arbitrary function with
a certain degree of continuity can be made lo satisfy certain boundary condition
s if
one wishes lo do so. In ihc principle of virtual displacements for instance, arb
itrary
functions of this type arc defined as virtual displacements but ihcy arc assumed
lo satisfy the homogeneous version of the displacement boundary conditions, i.e.
they are set identically lo zero at any points where the displacements are presc
ribed
even if Ihosc displacements (represented by u) arc not set to zero, i.e. w = 0 o
n
ihc parts of the boundary where u is given. This is done in order to eliminate
lerms of the type w which give rise to a type of 'work' one does not wish to
Idx J
have. In general however one can assume that \v and dw/dx can have values
different from zero on the boundaries and (his makes expression A.4) more genera
l.
The concept of an arbitrary function ?> used as a distribution function is relat
ed
not only lo virtual functions and consequently to virtual work but also lo the i
dea
of Lagrangian multipliers. These are functions of the vv type defined in order t
o
satisfy certain equations. They will be defined bellcr in what follows.
Although equation A.4) gives ihc user an insight into (he type of boundary
conditions required to solve the problem, these conditions have not yet been
explicitly incorporated into Ihc problem. In order lo do so let us consider thai
the
16
Chapter I Basic Concepts
boundary conditions are us follows:
? = it at .v = 0
du
,
A.6)
q = = q at x = I
r/.v
where the derivatives of ? are now defined as q and the terms with bars represen
t
known values of the function and its derivatives. It is usual to call the first
type
of conditions in A.6) 'essential' and those like q involving derivatives as 'nat
ural'.
Substituting those values into A.4) gives
It is now interesting to try to return to the original expression A.2) by
integrating by parts again, but this time passing the derivatives for v to u. The
first integration gives,
' [ du dw
<> ( dxtlx
u
? dw~\ r- -, ? - ? f/l ? <'"
L \Jv = 0
L ''vJ.v=i L rf-v
=0
A.8)
Notice that only the lerni in f? </iv/</.\:]? , disappears.
Furthermore the following expression results after carrying out a second
integration.
' Ur-u ,., , 1 , [du 1 [du 1 ? dwl
=0
A.9)
Once again only one term disappears, in this case [t/n'l^^o Notice that
q du/dx as defined earlier. - Grouping the terms together one now arrives at an
interesting expression, different from the original formula A.4) i.e.
} I'''" w + U2? - ?)? dx - [(</ - </) !. =, + ?(? - ?)''"' 1 ^=0 A.10)
? Ids2 J
L
</vJ.v = o
This expression implies that one is trying to enforce not only satisfaction of t
he
1.2. The Poisson's Equation
17
differential equation in v but the two boundary conditions. The w and dw/dx
functions can be seen as Lagrangian multipliers.
Furthermore nothing has yet been said about approximations; the above
expressions are valid for exact solutions as well. In other words the procedure
describes a general tool for the investigation of differential equations.
1.2 The Poisson's Equation
An important equation in engineering analysis is the so-called Poisson equation
which for two dimensions can be written as
+ =/ inQ
A.11)
f'.Yt P.Y]
or
V2u = h inQ
A.12)
<i2( ) ?2( )
where V2( ) = + , is called the Laplace operator, .*, and x2 are the two
coordinates and b is a known function of v,, v,. ? is the domain on which the
equation applies and is assumed to be bounded by ?. The outward normal to the
boundary is defined as n (figure I.I).
The Poisson equation or its homogeneous form (i.e. b = 0) which is the Laplace
equation, governs many types of engineering problems, such as seepage and aquife
r
analysis, heat conduction, diffusion processes, torsion, fluid motion and others
.
Consequently it is a very important equation in engineering analysis.
Figure 1.1 Domain under Consideration for Poisson Equation Basic Definitions
18
Chapter 1 Basic Concepts
Here one can also introduce the idea of multiplying equation A.12) by an
arbitrary w function, continuous up to the second derivative. This gives,
J (V2u - b)w da = 0
A.13)
n
Integrating by parts the terms in .v, and x2 gives
, / du dw du dw \
. du
Jj
bw)da+ \ wdV = 0
A.14)
fi \ dx% dxt dx2 dx2 / r dn
In this case the integration by parts of the two terms produces the derivative
of ? with respect to the normal, i.e. du/dn which will later on be called c/, i.
e.
q = ??/??.
Integrating by parts again, one obtains,
or
I (I.-' u + (- "' ? - bw) da + $* " w dr - f "V </? = 0
A. 15)
n \Axf Ax I
/
r dn
? ??
| {(V2w)h - bw} da + J'' - w </? - J ?(W dr = Q A.16)
n
???
? dn
Expression A.16) is equal to A.13) and hence one can write,
f (V2)w</fi = f (V2v)t/ </-f J'" ?-/?-f-?' " i/? = 0
A.17)
a
ii
iAi
? ?
where the term in b has been eliminated as it appears on the two sides of the
equation.
Equation A.17) can also be expressed in the form known as the Green's
theorem, i.e.
f {(V2u)n'-(V2tt')M}</fl= f ( '"w-i/ }</r
A.18)
h
? \?| dn/
Although this theorem is in many cases given as the starling point for many
engineering applications, including boundary clement formulations, it is much
more illuminating to use the concept of distribution as it illustrates the degre
e of
continuity required of the functions and the importance of the accurate treatmen
t
of the boundary conditions. In this regard let us now consider that the ? bounda
ry
of the a domain under study is divided into two parls, ?, and ?2 (?= ?, + ?2)
such that,
? = ? on ?,
<i.w>
du
q = -.. = q on ?2
dn
1.2. The Poisson's Equation
19
Hence equation A.16) can now be written as,
J {(V2u')tj-/m-; JO.
?
+ J qw</? + J qwdV- J ?'W df - J ? ilV = 0
A.20)
?,
l-j
?, <Vl ?; ??
Once again one can integrate by parts to retrieve the original Laplacian V2w
in order to see how the importance of the boundary conditions affect the equatio
n.
Integrating by parts once we have,
ii [ rlv, dxi i?x2 (lx2
+ f 'W ? dV + \ qw dV + f qw dr - J ii' " dV - J u\ dT = 0 A.21)
i ?
ii
i'j
i! nt i j r/i
One can split (he llrsl integral on ? into two terms (one on ?, and (he other on
?2), the second of which can be cancelled with the last integral in A.21). This
gives
J <
hw>dCl
a I dXi d%i dx2 dx2 J
+ J VWudr+ f qwdr+ | qwdr- f '^^ = 0 A.22)
?, Oil I ,
|'j
?, Oil
Integrating again by parts the following expression is obtained
I \{V2u)w - bw\ da
f wq ?/? + J ? </? + f qw ?/? + J qw dt
-f df = 0
A.23)
r, en
The first integral in ? can again be written as a summation of two integrals, on
e
on ?, and the other on ?2. The one on ?, can be cancelled with the integral on ?
,
of qw in Ihc above equation. This gives
J I (V2)u- - bw} du - J ? dV + j '" ? </? + J qw df
?
r2
?, ?| i-j
~1?(?\1? = 0
A.24)
20
Chapter I Basic Concepts
This P .la can be written as,
J J (V2h - b \w\ </Q - J (if - q)w A? + J (m - h) ' "' </? = 0 (! .25)
Once again this expression shows that one is trying to satisfy a differential
equation in the domain plus two types of boundary conditions, the 'essential*
conditions ? ? on ?, plus the 'natural' conditions q = </ on ?,. This is very mu
ch
what has been shown in equation A.10) with the only exception that the sign of
the last term is different in both expressions. This is because in A.10) the der
ivatives
were taken with respect to x rather than with respect to the normal, as they are
now.
1.3 Approximate Solutions
Although the previous sections have introduced the concept of distributions, the
formulations apply irrespective of the type of solution one finds, i.e. they are
valid
for exact as well as approximate solutions. This section however will investigat
e
what happens when the concept of an approximate solution is introduced in the
formulation. In engineering practice the exact solution can only be known in a f
ew
simple cases and it is hence important to see how the solution behaves when one
introduces an approximation. Let us consider now that the function ? defines an
approximate rather than the exact solution. In this case one can write for insta
nce,
? = ?1?1 +?2?2 +
A.26)
where x, are unknown coefficients and the ?, are a set of linearly independent
functions which are known, a, are generalized coefficients although in some
cases they can be associated with nodal values of (he variable under considerati
on.
In general in engineering problems, one prefers to use nodal values as they have
a clear physical meaning and this is done in finite elements, finite differences
or
the boundary element method. In such cases the approximation for 11 can be
written as
? = ? Ufa
i-\
where ?; are a set of linearly independent functions which arc sometimes called
interpolation functions. y are the nodal values of the field variable or its deri
vative
(or more generally the nodal value of any variable with physical meaning directl
y
related to ? or its derivatives).
Introducing the approximation for ? into the governing differential equation
one finds that the equation is no longer identically satisfied except for the ca
se in
which A.26) or A.27) can represent the exact solution. This produces an error or
residual function which will soon be defined.
1.3. Approximate Solutions
21
For instance, introducing an approximate value of i/ into equation (I.I) one
generally finds (hat
~" 3b*0 in.v
A.28)
Thc same will generally occur with the boundary conditions corresponding to this
equation, i.e.
? - ? ? 0 at ? = ?
A.29)
? - ij ? 0 at v = !
One can now introduce the concept of an error function or residual which
represents the errors occurring in the domain or on the boundary due to nonsatisfaction of the above equations. The error function in the domain is called
R
and is given by
R= ' + /?-?
A.30)
ilxand on the boundary one has,
ii
R, -?
and
A.31)
Although the above case is a particular and relatively simple equation the same
occurs for any oilier problem. If one considers the Poisson's equation A.12) for
instance, the error function in the domain is
R=V2u-b inQ
A.32)
and the errors for the boundary conditions (equation A.19)) are defined by
R i = it on ?,
K2 = ? ? ?? 12
The numerical methods used in engineering try to reduce these errors to a
minimum by applying different techniques. This reduction is carried out by forci
ng
the errors to be zero at certain points, regions or in a mean sense. This operat
ion
can be generally interpreted as distributing these errors. The way in which this
distribution is carried out produces different types of error distribution techn
iques
which, in general, force the integrals of the residuals weighted by a certain fu
nction
to be zero. Because of this they are called weighted residual techniques.
22
Chapter I Basic Concepts
1.4 Weighted Residual Techniques
The solution of the boundary value problem defined by equations A.28) and A.29),
A.32) and A.33) or similar sets for other problems can be attempted by choosing
an approximation for the function u. One can then have three types of method:
(i) If the assumed approximate solution identically satisfies all boundary
conditions but not the governing equations in Q. one has a purely 'domain'
method,
(ii) If the approximate solution satisfies the field or governing equations but
not the boundary conditions one has a 'boundary* method,
(iii) If the assumed solution satisfies neither (he field equation nor the bound
ary
conditions, one has a 'mixed' method.
Let us first assume that the functions </>, which arc defined to approximate ,
satisfy all boundary conditions. One then has a residual R function in the domai
n
as the field equations arc generally not identically satisfied. The idea is now
to
make R as small as possible by setting its weighted residual equal lo zero for
various values of (he weighting functions, i//,, such (hat
jKi^</Q = 0 inQ y=l,2,...,tf
A.34)
These functions have lo be linearly independent.
Notice that another way of writing A.34) in a form that is more compact and
easy to operate with, is by defining a new function n\ such that
H' - /W, + ???? +
+ Ms = t
where fi} arc arbitrary coefficients. Hence equation A.34) can now be written in
a more compact form as,
[RwJQ = O in ?2 A.36)
Different types of weighting functions \j/} (or w) will define different approxi
mate
methods. Equation A.34) or A.35) will produce a system of algebraic equations
from which the unknown values of the a, or u, coefficients used in ? (equation
A.26) or A.27)) can be obtained.
The approximation can always be improved by increasing the number of N
functions used. (N is the number of terms in the approximate solution equal lo
the number of weighting functions required.)
Approximate methods based on equation A.36) arc called weighted residual
methods and, given an approximate solution, the method will vary in accordance
with the functions used as weighting functions. In what follows a few will be
reviewed.
1.4. Weighted Residual Techniques
23
(i) Subdomain Collocation
For this method the domain Q is divided in M subdomains and the integral of
the error in each of them is set to zero. The weighted functions arc simply chos
en as.
@ for .? ? ilj
(? indicates belonging to and ilj is the '/' subdomain). liquation A.34) becomes
.
fR</.v = O; ./=1,2, ...,N
A.38)
(ii) Galcrkin Method
In the case of Galerkin's method the weighting functions are the same as the
appoximating functions, i.e.
?]-?) (I.?? I
hence equation A.34) becomes,
| ??,A? = 0 /=l,2, ...,N
A.40)
ii
Using the same definition as in A.35) this can be written as,
it
with,
iv = /f,f/>, + 112?2 + ... + /V/'iv
(I-42
This method is the starting point of many finite clement formulations for which
the symmetry of ?} = ?} coupled to inherently symmetric field equations, lead to
symmetric algebraic matrices.
(Hi) Point Collocation Method
In this case N points .Vi,.v2... .,.v,v arc chosen in the domain and the residua
l is
set to zero at these points This operation can be interpreted as defining weight
ing
functions in terms of Dirac deltas, i.e.
J R dx = J [flf i Fx
4) + a2Fx
2) + x] dx = 0
1/2
1/2
which produce the following system of equations
-1.2a, -0.25a2 + 0.125 = 0
(i)
0.25a,+ 1.2a2 +0.375 = 0
from which one can obtain,
Substituting (j) into (e) gives the following result
x x3
Notice that the exact solution (c) has been obtained since the assumed shapes of
? are able to represent it.
26
Chapter 1 Basic Concepts
(H) Galcrkin
In this case the weighting functions are,
?1=?1
and the weighted residual expressions arc
f [a, Fx - 4) + a2Fx - 2) + x](x - 2.x2 + x3} dx = 0
:
f [a,Fx - 4) + 2Fx - 2) + .x](x3 - x2) dx = 0
?
which produces the following algebraic equations in a, and a2.
-4,+2+l=0
(m)
a, -4a2-1.5 = 0
This also results in
(iii) Point Collocation
Here one forces the residual to be zero at a series of points. Consider in this
case
that R is zero at the two points x = 0.25 and x = 0.75. This gives
K|*=o.2s I0a,2a2+I0
(n)
ix=o.2s = 2a, + I0a2 + 3 = 0
with the same results for <xx and a2, i.e.
Notice that this case is rather trivial and the same results have been obtained
for all the methods. In general this will not be true when the exact solution ca
nnot
be reproduced by the proposed value of ? and one will find different results
depending on the method used.
Example 1.2
Let us now study another equation using point collocation such that in this case
we will obtain an approximate rather than the exact solution.
1.4. Weighted Residual Techniques
27
Consider the equation (I.I), with ?2= I and .v = b, i.e.
, +u+x=0
(a)
ax
and Ihe homogeneous boundary conditions, ? = 0 at x = 0 and x == I.
The exact solution of (a) can be easily obtained by integration and gives
=
x
(b)
sin I
Instead of using (b) we will try to approximate it defining a solution
? = ?1?1+???2 + ?3?? +...
(?)
wiiere the ?/ are terms of a polynomial in .v, i.e.
0, = l, ?? = ?, </.., = .y2... (d)
In order to satisfy the boundary conditions exactly, equation (c) has to give,
? = 0 at .x = 0 and x = I
(e)
which implies that,
at .v = 0-m = . =0
(f)
at x = I - u = at + a2 + ?3 + ... = 0
Hence a t = Oand a2 can be expressed in function of the other, parameter, i.e.
?? =
(a3 + a4 + ...) (g)
Substituting a, =0 and (g) into (c) one can write,
? = a3{x2 - x) + a4(*3 - *) + aS(*4 - x) + ...
()
= x(l -x)(-a3 - a4) + x(l -x)(-a4)x + ...
Defining now a new set of unknown parameters a, such that,
at = -a3-a4; a2=-a4... (i)
one can write,
+x2x + ...)
(j)
28
Chapter 1 Basic Concepts
This function satisfies the boundary conditions in ? and has the degree of
continuity required by the derivatives in equation (a), hence it is said to be
'admissible'. We will also see that the 'distance' between the approximate and
exact solution decreases when the number of terms in (j) increases and this impl
ies
that the approximate formulation ? is 'complete', i.e. tends to represent the ex
act
solution better and better when the number of terms increases.
In order to apply the point collocation technique we will restrict ourselves to
two terms in the (j) expression, i.e.
? = .?A-?)(*,+?2?)
(k)
Substituting this function into the governing equation (a) one finds the followi
ng
residual, i.e.
d2u
tlx1
R =l.." + ? + x = (-2 + x - x2)a, + B - 6v + x2 -x3)a2 + x
(I)
I.
Collocation can now be interpreted as setting R =0 at two points, say x
| and
.v = > This can also be expressed in terms of Dirac delta functions applied at th
ese
two points, i.e. the weighting function is,
iv = /JlAl(.x-i) + /f2A2(.x-J) (m)
The weighted residual integrals are represented by
jKv(/x = 0
(n)
?
or simply,
R = 0 at x = Jand x = 2 (o)
Substituting these values of x into (I) one obtains two equations in a, and a2.
They can be written in matrix form as follows,
??::?;}
The solution of this system gives
ai = sV *2 = 24t7
(?)
The approximate value of ? - equation {k) - can now be written as.
D2 + 0?)
(r)
1.4. Weighted Residual Techniques
29
Table 1.1 Results for Point Collocation
.X
0.10
0.30
0.50
0.70
0.90
? (exact)
0.018641
0.051194
0.069746
0.065585
?.???9?1
? (approximate)
0.019078
0.052258
0.071428
0.065806
0.032350
R
-0.009953
+0.002027
+0.00000
-0.024884
-0.081474
Notice that the error function can now also be fully defined in terms of x, by
substituting a, and ?, into A). This gives,
? = ,;7(-4+ I9.x-2.x2-40.xJ)
(s)
Those results can be tubulated in table 1.1 where they are compared against the
exact solution for u. Notice that the values of R are identically zero at x = |
and
v = , but that this does not mean that the solution for u is exact at those poin
ts.
K\amplc 1.3
Let us apply Gulerkin's technique to equation A.1) for which "/} = 1 and b = x
with homogeneous boundary conditions ? = 0 at .x = 0 and .x = 1. The approximate
solution will be the same as in example 1.2., i.e.
ot2.x2(l-.x)
(?)
which can be written as
!/ = !)(,</>, +<X2</>2 (?)
where ?? and ?2 are the shape functions (?, -?(\ ?); ?2 = ?2A ?)). The
residual is the same as previously, i.e.
R- ~+u+x
dx2
(c)
- (- 2 + .x - .x2)a, + B - 6.x + .x2 - .x3)a2 + .x
The weighting function iv in Galerkin is assumed to have the same shape function
as the approximate solution (b), i.e.
*- = /*,?1 + /^2
(d)
The coefficients //, and fi2 are arbitrary.
30
Chapter I Basic Concepts
The weighted residual statement is,
t
$Rwdx = 0
(c)
?
which produces two integral expressions as /?, and /72 are arbitrary, i.e.
dx = O @
?
or simply,
& = () and |??2?? = 0 (g)
?
?
Substituting (c) and the functions </>, and ?2 into (g) gives
f [(-2 + x - .v2)a, + B - 6? + ?2 - ??)??2 + ?][?< I - ?)] d.\ = 0
?
(h)
l
J [(-2 + x - x2)! + B - 6x + x2 - x3)a2 + x]j>2(l - x)] dx = 0
?
After integration this gives the following system
La mJuJ ?
Notice that the matrix is symmetric because the equation is of an even order and
the approximate and weighting functions arc the same. Solving (i) gives
Substituting these values into (a) produces the approximate solution for u, i.e.
?*=?A-?)(?^ + ^?)
(?)
One can also find the residual function R (equation c) which is now
R = 369 {- 16 + 62\ - 8x2 - 63xJ)
The results for ? and R are given in table 1.2 where they arc compared against
the exact solution of u. Notice that although the solution is overall more accur
ate
than in the case of using the collocation technique, one now needs to carry out
. some integrations as shown in formula (h). This operation was not required for
the case of point collocation.
1.5. Weak Formulations
Table 1.2 Results for Galerkin
31
AC
0.1
0.3
0.5
0.7
0.9
? (exact)
0.018641
0.051194
0.069746
0.065582
O.O3O9OI
? (approximate)
0.018853
0.051162
0.069444
0.065505
0.031146
R
-0,026945
+0.000485
+0.013888
+ 0.005070
-0.034165
1.5 Weak Formulations
The fundamental integral statements of the boundary element and the finite eleme
nt
methods can be interpreted as a combination of a weighted residual statement
and a process of integration by parts that reduces or 'weakens' the order of the
continuity required for the ? function.
If one returns to equation A.12) with b = 0 for simplicity, i.e.
V2m = 0 in ?
one can write formula A.25) as,
dw
J (V2m)v </fi - | (q - q)w dV + J (? - ?) </? = ?
U
?2
I" i
f'"
or in terms of residual functions.
A.47)
A.48)
f Rwdu- J Rzw
?, '"
A.49)
A special case of this equation is the case for which the function ? exactly
de
integral, one finds,
J (V2u)" </Q = - J cjw</?- J qw dV + J ?' "'dT + f ? ' "' A? A.57)
n
??
r,
r, On ij ??
1.5. Weak Formulations 33
This is i he starting statement for the Boundary Plement formulation of the Lapl
ace
equation. The same equation can be obtained starling from the integral of the
weighted residual over the domain Q (equation A.53)), integrating by parts twice
and then introducing the boundary conditions. The processes have^tready been
shown from another Field equation in formulae A.13) to A.16) and then A.19) and
A.20), the only difference now being that h is zero.
Consider now equation A.1) again to illustrate how a weak formulation can
be used and the domain and boundary element statements are obtained. Let us
start with equation A.10) which was deduced from A.1) by a process of integratio
ns
by parts and application of boundary conditions, i.e.
} \d~" w + (/.2? - ?)? dx - L(q - qW]xm,+[(- ?) ^] = 0 A.58)
i ??- J
L *_U=o
which can also be expressed in a more compact form in function of residuals, i.e
.
jKuw/v-[R,vv]v=,+ K,'~ =0
A.59)
L <f*_Uo
<
The function ? will now be assumed to satisfy exactly the 'essential' boundary
conditions it = ? at x = 0. In this case A.58) becomes,
f '"" w + U2u - b)w dx = l(q - M*-1
? ??- J
A -60)
J
or in terms of A.59), simply
i
j R\\dx = llt1wl^l
A.61)
' <
Integrating by parts equation A.60) one can write
'" ('1V + (??? _ h)J dx = j- -, _ ,- -, A 62)
x dx
J
If the weighting function w is forced to satisfy the homogeneous version of the
essential boundary conditions at x = 0, equation A.62) becomes,
\\? + &uh)U}jxtqw}x.l A.63)
? (. dx dx
J
which is analogous to equation A.55) obtained for the Laplace field equation.
Notice that equation A.59) can also be obtained by applying the boundary
conditions into statement A.3) and that this statement was simply obtained by
integrating by parts weighted residual expression A.2).
34
Chapter 1 Basic Concepts
The Boundary Element type governing statement for the example under
discussion is found by carrying out two consecutive integrations by parts of A.1
0)
and this gives the previously obtained formula A.7), i.e.
i f j2w -j
j p + (?2? -b)w\ dx
1L k_L-i L
This expression could also have been obtained by carrying out a double
integration by parts of the weighted residual equation A.2) and applying afterwa
rds
the boundary conditions.
It is worth noting that both in this one dimensional example and the two
dimensional Laplace equations, a Finite Element type statement has been obtained
after the first integration by parts (equations A.3) and A.14)), and Boundary
Element type integral equation after the second integration (equations A.4) and
A.15)).
Example 1.4
In order to understand the effect of weak formulations on the satisfaction of
boundary conditions, we will now consider again equation (I.I) but assume that
the boundary conditions are of two types, i.e.
at x = 0 - ? = 0
('essential' condition)
du
?. . (?)
atx=l-*q = = q (natural condition)
dx
The expression previously used for the approximate values of ? can not now
be applied as the boundary conditions arc different. Let us consider again the
starting expression,
? = ?1+?2? + ?3?2 + ... (b)
and satisfy exactly the essential condition, at .v = 0, i.e.
but not the natural condition.
Hence the approximate solution is now,
M = a,.x + a2.v2 + ... (d)
where a, = ?2 ?? = lh
1.5. Weak Formulations 35
The residual will now be different from the one in the previous examples, i.e.
R
+
+u + x = <x,x + oi2B + x) + x (e)
dx1
The weighted residual statement has to include now the natural boundary
condition R2 residual which is not identically satisfied, i.e.
|Rwd.x = [R2vvL = l
(f)
?
or in expanded form,
\
(g)
! (d2u \
J I +U + XJW dx = l(q - ij)wjx,,
One can now solve equation (g) in its present form or reduce the order of deriva
tives
in the domain and the number of terms on the right hand side by integrating by
parts the d2u/Ux1 term. This gives
J ?" ^ - (" + x)w\dx =
? {dx dx
)
(h)
Notice that in Galcrkin the weighting function iv has the same shapes as the
approximation for ? (equation (d)). Hence for two terms,
and
where
?% x and ?? = ?2.
Substituting these values into (h) one finds.
- (<X ,</.,+ <?2?2 + X )(fi, ?, + A
2?2 ) I dx
or
} {(a, + 2a2.*M/, + 2/2.v)- (,.v + x2x2 + .v)(/i,.v + fl2x2)} dx
?
36
Chapter ! Basic Concepts
As the /I, and (!2 terms are arbitrary this implies satisfaction of the followin
g
two equations,
i
f {(a, + 2oc2x) - (a,x2 + a2x3 + x2)] dx = q
?
(I)
i
dx1
A solution of equation (i) regardless of boundary conditions is
f
,. x*Xl (j)
X;,
.X > X;
Once the boundary conditions are applied, equation A.72) becomes
m, + | xiv </.x + </, w,
q0w0 = 0 (k)
?
Taking into consideration that ?'? = 0 and substituting the other values of iv a
s
given by (j) one finds,
x,
I
?, = J x2 dx + f x,,x dx + qtXi (I)
?
xi
or,
Vi Xf
Notice that only one unknown (</,) remains, since one of the boundary stresses
disappeared because of the variation of the weighting function w. The value of <
/,
can be determined by taking the coordinate .?( = I, i.e.
which is the exact value in this case.
Any value of ? inside the domain can be computed from A), i.e.
1/2 = 48
(O)
which is also the exact solution.
If instead of the fundamental solution given by (j) one had chosen a fundafundamental solution that also satisfies the boundary conditions, then no unknow
ns
would exist either in the domain or at the boundaries and the value of ? at any
point would be obtained by a single integration. Consider for instance the solut
ion,
_|A-.??? .?^.?,"'= . '
(p)
1.6. Boundary and Domain Solutions
41
This fund ion satisfies ? and (he boundary conditions w = 0 at .x = 0 and .x = 1
(?1,, = ?1, =0). Thus equation (k) gives
I
Aj
1
x)x dx (q)
I,- = J xw dx = f (I - .x,).x2 (/.x + J x,(\
?
0
?.
resulting in
->?
which is the exact solution.
Fundamental solutions that satisfy the boundary conditions as well as the
governing equations are called Green's functions.
(iii) Domain Solution
The weighted residual statement used here is the one resulting after one integra
tion
by parts has been carried out (equation A.73)), i.e.
' f du dw ) Vdu T
H-: . +xw\dx + \Tw\ =0 (s)
? (. dx dx J
U* Jo
The proposed approximate solution is the one in Example 1.1 which satisfies the
boundary conditions, i.e.
? = ?1^,+?2?2 (t)
with
?% x
2.x2 + .x3
??4?+1
dx
= Xx'
2.x
dx
) = 0
)-.v(I -x)
using a trial function of the form u = A{y)x{x
1) and point collocation for .v 1
/2,
0 < ? < X) as collocation point.
1.4.
Solve exercise 1.3 using the Galcrkin method with the integral and weigh
ting function
only along the ? axis.
1.5.
Solve the equation I (I + u) =0 from x = 0 to x = 1 with boundary condit
ions
dx L
.</? J
n@) = 0 and u(l)=l using ? ao + avx-\-?2\? as trial function and subdomain
collocation with only one subdomain (? from 0 to I).
1.6.
The equation of the vertical displacement of a cable suspended between t
wo points
dzu
is + p(x) = 0 where p{x) is the ratio between the distributed load and the horiz
ontal
dx*
force at the extremes.
Use the weak formulation and the homogeneous approach of boundary solution
to compute the slope at the extremes for a cable that extends from .\ = 0 to ,v=
I
with boundary conditions (()) = u(l) = (). The function p(x) is given by
/>(.v) = 0
?(?)=1
p(x) 0
0 < x <
1/4 < x
3/4 ^.v
1/4
<3/4
<|
1.7. Using the same equation of exercise 1.6 and the singular approach of bounda
ry
.solution compute the value of ? at the mid-point.
Chapter 1 Basic Concepts
The same us exercise 1.7 using a fundamental solution that also satisfies the bo
undary
conditions (Green's function).
Solve the same equation of exercise 1.6 by means of a domain solution procedure
and Galerkin. Use the following approximate solution
? = t s'n k.v + 2 sin ???
Chapter 2
Potential Problems
2.1 Introduction
Since the publication of the first book on Boundary Elements in 1978 [1] many
such works have appeared in the literature, some dealing with potential, others
with clastostutics problems as described in Chapter 3 and many other engineering
applications. The importance of 1978 is that on that date publication of the Fir
st
book on boundary elements coincided with holding the first conference in which
Ihe method was established, although the name in the context of the now classica
l
B.E.M. appears to have been used for the first time in two papers by Brebbia
and Dominguez and dated 1977 [2], [3]. Up to that time boundary integral
equations solutions were almost exclusively the domain of mathematicians and
physicists, with very little work being done to apply them to realistic engineer
ing
problems. Efforts such as the pioneering work by Hess and Smith [4] remained as
special cases rather than being interpreted as a way of generating a whole new
method of solutions for general engineering problems. Nevertheless Hess and Smit
h
developed many powerful programs for the solution of Laplace type boundaryvalue problems which were applied to potential flow and arbitrary bodies, using
what is now called the indirect boundary element technique. They extended their
formulation to analyse three dimensional objects as well as two dimensional and
their codes arc still popular in aerodynamics. Equally important although in a
different field was the work of Harrington and his collaborators [5], [6] who
applied the technique to solve electrical engineering problems using more genera
l
impedance boundary conditions, of the type now called Robin or mixed conditions.
They were still using the indirect formulation.
Boundary elements are usually associated with the direct formulation which
is the one discussed most of the time in this book. This formulation in potentia
l
problems can be traced to Jaswon. As early as 1963 Jaswon [7] and Symm [8]
presented a numerical technique to solve Fredholm boundary integral equations,
which consisted of discretizing the boundary into a series of small segments
(elements) and assuming a constant source density within each segment. They
employed collocation to obtain the governing system of equations and compute
the influence coefficients using numerical - Simpson rule - techniques, with the
exception of the singular coefficients which were computed either analytically o
r
by the summation of off-diagonal terms. They even proposed a more general
formulation through the application of Green's third identity with potentials an
d
their derivatives as boundary unknowns and results for this formulation were
represented in [7] and [9]. All the bases of boundary elements were there but
46
Chapter 2 Potential Problems
somehow their work failed to attract the attention it deserved, probably due to
the simultaneous emergence of the finite element method.
Since 1978 the boundary element method is seen as related to other numerical
techniques such as finite elements and finite differences, mainly through the wo
rk
of Brebbia and his collaborators. This relationship is sometimes highlighted by
using weighted residual or variational type techniques.
In this chapter the formulation of boundary elements for potential problems
will be discussed. The method is presented using the same consideration of weigh
ted
residuals as used in Chapter 1. The numerical implementation of the method will
be described in detail, and used in simple computer codes which can be run in
the type of PC used by engineers. The chapter presents formulations using differ
ent
types of elements, i.e. those with constant and linear variations. Special consi
derconsideration is given to the proper treatment of the corner points. Quadratic e
lements
are also discussed and implemented in a computer code.
Other sections in this chapter deal with the Poisson equation and the treatment
of distributed sources, problems with more than one surface, multizoncs applicaapplications, anisotropy and the Helmholtz equations. Computer codes arc not pro
vided
for these cases but the existing programs can be easily extended and the boundar
y
clement student can do this as a computational exercise.
boundary conditions are applied. Our aim is now to render formula B.8) into a
boundary integral equation. This is done by using a special type of weighting
function u* called the fundamental solution.
Fundamental Solution
The fundamental solution ?* satisfies Laplace's equation and represents the fiel
d
generated by a concentrated unit charge acting at a point 7'. The effect of this
charge is propagated from i to infinity without any consideration of boundary
conditions. Because of this the solution can be written
V2* + A' = 0 B.9)
where A1 represents a Dirac Delta function which tends to infinity at the point
x = x1 and is equal to zero anywhere else. The integral of A* however is equal t
o
one. The use of Dirac delta function is an elegant way of representing unit
concentrated charges as forces when dealing with differential equations.
The integral of a Dirac delta function multiplied by any other function is equal
to the value of the latter at the point ?1. Hence
f m(W) dQ = f ?(- A1) du = -?* B.10)
Equation B.8) can now be written as,
'+ J uq* dT+ J uq*dT= J qu*dr+ J qu* dT
B.11)
?2
Ti
??
?,
It needs to be remembered that equation B.11) applies for a concentrated charge
at T and consequently the values of u* and q* are those corresponding to that
particular position of the charge. For each other x1 position one will find a ne
w
integral equation.
For an isotropic three dimensional medium the fundamental solution of
equation B.9) is
?* = --B.12)
4nr
2.2. Basic Integral Equation
49
and for a two dimensional isotropic domain, it is
B.13)
2?
where r is the distance from the point x' of application of the delta function t
o
any point under consideration.
It is easy to check that solution B.12) and B.13) satisfy the three and two
dimensional Laplace equations. Consider for instance the three dimensional
equation in terms of polar coordinates after neglecting terms which are zero due
to symmetry of the solution, i.e.
VV->^ + 2/)"*=-A'
B.14)
or r vr
Simply by substituting solution B.12) into B.14) we can check that the equation
is satisfied for any value of r different from zero. For the case where r = 0 we
need
to carry out the integration around a sphere of radius ? and then take ? to zero
.
Consider that the sphere has an ft? domain, and integrate by parts to express th
e
Lapiacian in terms of boundary fluxes du*/i)n, i.e.
i <V2*) ?1 = J ?) ,/? = J *f dT
B.15)
Notice that n = r on the surface of the sphere.
Substituting now the fundamental solution B.12) into B.15) and making r (or
r.) tend to zero gives
{} j
-ihnf-4"V
c-0 ( 4ff?ZJ
B.16)
Notice that the surface of the sphere is ?? = 4??2. Similarly for the two
dimensional case one can define a small circle of radius e and then take the lim
it
when ? -0, i.e. .
{
f 2ne)
= hm{ >=-l
B.17)
Here the perimeter of the small circle is ?? = 2??.
50
Boundary Integral Equation
Chapter 2 Potential Problems
We have now deduced an equation B.11) which is valid for any point within the
Q domain. In boundary elements it is usually preferable for computational reason
s
to apply equation B.11) on the boundary and hence we need to find out whul
happens when the point x1 is on ?. A simple way to do this is to consider that
the point i is on the boundary but the domain itself is augmented by a hemispher
e
of radius e (in 3D) as shown in figure 2.2 (for 2D the same applies but we will
consider a semicircle instead). The point x1 is considered to be at the centre a
nd
then the radius ? is taken to zero. The point will then become a boundary point
and the resulting expression the specialization of B.11) for a point on ?. At pr
esent
we will only consider smooth surfaces as represented in figure 2.2 and discuss t
he
case of corners in other sections.
Boundary surface ?
(i) Three Dimensional Case. Hemisphere around i
Boundary curve rt
Boundary curve ?
Boundary point
(it) Two Dimensional Case. Semicircle around i
Figure 2.2 Boundary points for two and three dimensional case, augmented by
a small hemisphere or semicircle
2.2. Basic Integral Equation
51
It is important at (his stage to differentiate between two types of boundary
integrals in B.11) as the fundamental solution and its derivative behave differe
ntly.
Consider for the sake of simplicity equation B.11) before any boundary condition
s
have been applied, i.e.
?1? J uq*dY = J ?*</ A? B.18)
i
r
Here ? = ?, + ?2 and satisfaction of the boundary conditions will be left for la
ter on.
Integrals of the type shown on the right hand side of B.18) arc easy to deal
with as they present a lower order singularity, i.e. for three dimensional cases
the
integral around ?? gives:
t-o (.r,
= lim Uq ' dA
c-o (r, 4/re J
B.19)
In other words nothing occurs to the right hand side integral when B.11) or B.18
)
are taken to the boundary. The left hand side integral however behaves in a
different manner. Here we have around ?? the following result,
Vj
?,
Notice that we are assuming throughout that the fundamental solution is applied
at a particular '/' node, although this is not explicitly indicated in ?*, q* no
tation
to avoid proliferation of indexes. Hence for a particular '/' point one can writ
e,
\u'+ t /J'V-i C'V
B-26)
i=l
(=1
2.3. The Boundary Element Method
53
Nodes
element
element
(b) Linear Elements
Nodes
element
(c) Quadratic Elements
Figure 2.3 Different types of boundary elements
If we now assume that the position of i can also vary from 1 to JV, i.e. we
assume that the fundamental solution is applied at each node successively one
obtains a system of equations resulting from applying B.26) to each boundary
point in turn.
54
Chapter 2 Potential Problems
Let us now call
1?" + wheni=;
hence equation B.26) can now be written as
t HV= t C'V
B-28)
This set of equations can be expressed in matrix form as
HU = GQ B.29)
where H and G arc two /V x N matrices and U, Q arc vectors of length N.
Notice that /V, values of u and N2 values of q are known on ?, and ?2
respectively (?, + ?2 = ?), hence there are only N unknowns in the system of
equations B.29). To introduce these boundary conditions into B.29) one has to
rearrange the system by moving columns of H and G from one side to the other.
Once all unknowns are passed to the left-hand side one can write,
AX = F B.30)
where X is a vector of unknowns u's and q's boundary values. F is found by
multiplying the corresponding columns by the known values of u's or q's. It is
interesting to point out that the unknowns arc now a mixture of the potential an
d
its derivative, rather than the potential only as in Finite elements. This is a
consequence of the boundary element being a 'mixed' formulation and gives an
important advantage to the method over finite elements.
Equation B.30) can now be solved and all the boundary values arc then known.
Once this is done it is possible to calculate any internal value of u or its der
ivatives
The values of w's arc calculated at any internal point V using formula B.11) whi
ch
can be written as,
?1' = f qu* dr - J uq* dT
B.31)
?
?
Notice that now the fundamental solution is considered to be acting on an intern
al
point T and that all values of ? and q are already known. The process is then
one of integration (usually numerically). The same discretization is used for th
e
boundary integrals, i.e.
"'= I C'V- I W'V
B.32)
i-i
j=i
The coefficients G'7and Hu have been calculated anew for each different internal
point.
2.3. The Boundary Element Method
55
The values of internal fluxes in the two directions, say .*, and .v2, qXl = ??/?
??
and qxi = ??/??2, are calculated by carrying out derivatives on B.31), i.e.
?|/ ? \??,/
? \dxj B.33)
Notice that the derivatives are carried out only on the fundamental solution
functions ?* and u* as we arc computing the variations of flux around the'/' poi
nt.
The boundary integrals are discrctized into integrals along the elements.
The kernels to be integrated along the elements are
= ?-^~(-/??) =
rk
B.35)
*ty 2? cx4
2??
and
[Br2, - l), + 2r,r22] B.36)
2irr2
T~ = "^? lBr-2 " 1} + 2r'rdx2 2???2
where rk indicates derivative at the integration point; i.e.
'drs
rV_
and n,, n2 are the components of the unit normal. The integration of the expresexpressions given in B.36) is done numerically using a standard Gaussian quadrat
ure.
Evaluation of Integrals
Integrals like GiJ and fiij in the above expressions can be calculated using num
erical
integration formulae (such as Gauss quadrature rules) for the case i?=j. For the
element i = j however the presence on that element of the singularity due to the
fundamental solution requires a more accurate integration. For these integrals i
t
56
Chapter 2 Potential Problems
Node i <2>
'
?
112
|
- 1/2
r*Figure 2.4 Element coordinate system
is recommended to use higher-order integration rules or a special formula (such
as logarithmic and other transformations which will be discussed later on).
For the particular case of constant elements however the H" and G" integrals
can be computed analytically. The ?" terms for instance are identically zero, as
the normal n and the element coordinate are always perpendicular to each other,
i.e.
//" = f q* iir = f A"* ^ /? *5 0
B.37)
r,
n <V <"
The integrals in G" require special handling. For a two dimensional element for
instance they are
G" = J m* A? = 1 f ln(' )dr
B.38)
?,
2??. \r/
In order to integrate easily the above expression one can change coordinates to
a
homogeneous <!; coordinate over the element (figure 2.4) such that,
r =
B.39)
where / is the element length.
?
CALL GHHATPC(X,Y,XM,YH,0,H,FI.DF1,KODE,NX)
?
? SOLVE SYSTEM OF EQUATIONS
?
CALL SLNPD(G,DFI,D,N,NX>
?
? COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
CALL INTERPCIFJ,DFI.KODE.CX.CY,X,Y,POT,FLUXI,FLUX2)
?
? PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS
?
CALL OUTPTPC(XM,YM,F1,DFI,CX,CY,POT,FLUXI.FLUX2)
?
? CLOSE INPUT AND OUTPUT FILES
?
CLOSE (INP)
CLOSE (IPR)
STOP
END
60
Chapter 2 Potential Problems
Routine INPUTPC
All the input required by the program is read in the program INPUTPC and
contained in a file whose name is requested by the main program. The file should
contain the following input lines (using free FORMAT):
A)
Title Line One line containing the title of the problem.
B)
Basic Parameter Line One line containing the number of boundary
elements and the number of internal points where the function is required.
C)
Extreme Points of Boundary Elements Lines The coordinates of the extreme
of the elements read in counterclockwise direction for the case shown in
figure 2.6(a) and in clockwise direction for 2.6(b).
D)
Boundary Conditions Lines As many lines as nodes giving the values of
KODE and the value of the potential at the node if K.ODE =0 or the value
of the potential derivative if KODE = I.
E)
Internal Points Coordinates Lines The .x,x2 coordinates of the internal
points are read in free FORMAT in one or more lines.
This subroutine prints the title, the basic parameters the extreme points of the
boundary elements and the boundary conditions. The internal point coordinates
are printed in the OUTPTPC routine.
Notice that all input is written in free FORMAT.
SUBROUTINE INPUTPC(CX,CY,X,Y,KODE,FI)
?
? PROGRAM 2
?
CHARACTER*80 TITLE
COMMON N,L, INP.IPR
DIMENSION CXA),CYA),XA),YA),KODEA),FIA)
?
? ?" NUMBER OP BOUNDARY NODES ("NUMBER OF ELEMENTS)
? L= NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED
?
WRITE(IPR,100)
100 FORMATC ',79('*'))
?
? READ JOB TITLE
?
READ(INP,'(A)') TITLE
WRITE(IPR,'(A)') TITLE
?
? READ NUMBER OF BOUNDARY ELEMENTS AND INTERNAL POINTS
?
READ(INP,*)N,L
WRITE(IPR,300)N,L
300 FORMAT(//' DATA'//2X,'NUMBER OF BOUNDARY ELEMENTS =' ??/2?, 'NUMBER
1 OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED -',I3)
?
?
? READ COORDIKATES OF EXTREME POINTS OF THE BOUNDARY ELEMENTS
? IN ARRAYS X AND Y
?
WRITE(IPR,500)
500 FORMAT(//2X, 'COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELE
1MENTS',//IX,'POINT',7X,'X',15X, 'Y')
READ(INP,*) (X(I),Y(I),I-1,N)
DO 10 I-1,N
10 WRITE(IPR,700I,X(I),Y{I)
700 FORMATBX,I3,2BX,E14.5))
2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 61
(a)
Numbering direction for external surface
CLOSE DOMAIN (anticlockwise)
OPEN
(b) Numbering direction for internal surface
DOMAIN (clockwise)
Figure 2.6 Numbering directions for external and internal surfaces
? READ BOUNDARY CONDITIONS IN FI(I) VECTOR, IF KODE(I)0 THE FI(I)
? VALUE IS A KNOWN POTENTIAL;IF KODE(I)1 THE FI(I) VALUE IS A
? KNOWN POTENTIAL DERIVATIVE (FLUX) .
?
WRITE(IPR,800)
800 F0RMAT(//2X, 'BOUNDARY CONDITIONS'//2X, 'NODE' , 6X, 'CODE' , 7X, 'PRESCRI
1BED VALUE')
DO 20 1=1,N
READ(INP,*) KODE(I),FI(I)
20 WRITE(IPH,950I,KODE(I),FI(I)
950 FORMATBX,I3,9X,I1,8X,E14.5)
?
? READ COORDINATES OF THE INTERNAL POINTS
?
IF(L.EQ.O) GO TO 30
RKAD(INP,)
30 RETURN
END
Routine GHMATPC
The routine GHMATPC forms the G and H matrices of equation B.29) through
its subroutines EXTINPC and LOCINPC. It then rearranges their columns to
form A matrix and F vector of B.30).
The subroutines EXTINPC and LOCINPC perform the following functions.
EXTINPC: This subroutine computes the H and G matrix elements by means
of numerical integration along the boundary elements (using 4
points Gauss quadrature).
It calculates all elements except those on the diagonal.
LOCINPC: Only calculates the diagonal elements of G matrix, given by
equation B.41),
62
Chapter 2 Potential Problems
Notice that as ?? = 0 the diagonal elements ?? ? are simply \.
It is also important to point out that as the fundamental solution has been
RETURN
END
2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 63
Routine EXTINPC
This subroutine computes the values of the off-diagonal coefficient of H and G
using a 4-point Gauss integration formula (see Appendix A). It also computes,
using the same numerical integration formula, the integrals of the fundamental
solution and its derivatives required for the computation of potentials and flux
es
at internal points (equations 2.35 and 2.36).
Consider now that instead of the system a, .v2 we use an x-y system of
coordinates. In this case A'I, X2, Y\, Y2 are going to be the coordinates of the
extreme points of each element considering them in clockwise (open domain) or
anticlockwise manner (closed domain).
Using numerical integration and changing to a dimensionless system of
coordinates the GiJ and HiJ terms for each element and collocation point can be
written as,
+ (Yl-Y2f
(RA)k
//' '= > In ]wk
= V __._!_.. (RDI*ETA1 +RD2*ETA2)
(RA)
where
(Y\ YP
??-??--(? YP
? is the weighting for each point, XP, YP are the coordinates of the collocation
point and ETA1, ETA2 are the components of the unit normal. The values wk
and the location of the ? points over the element are given in Appendix A.
c
SUBROUTINE EXTINPC(XP,YP,X1,Y1,X2.Y2,H,G,I>1,D42,DU1,DU2,K)
?
? PROGRAM 4
?
? THIS SUBROUTINE COMPUTES THE INTEGRAL OF SEVERAL NOH-SIUOULAR
? FUNCTIONS ALONG THE BOUNDARY ELEMENTS USING A FOUR POINTS
?
GAUSS QUADRATURE
?
WHEN X=0, THE OFF DIAGONAL COEFFICIENTS OF THE ? AND G MATRICES ARE
?
COMPUTED
?
VHEN Kl, ALL THE COEFFICIENTS NEEDED FOR COMPUTATION OF THE POTENTIAL
?
AND FLUXES AT INTERNAL POINTS ARE COMPUTED (G,H,E1,E2,Fl,F2)
?
?
?
RA* RADIUS DISTANCE FROM THE COLLOCATION POINT TO THE
?
GAUSS INTEGRATION POINTS ON THE BOUNDARY ELEMENT
?
???1,???2
COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
?
RD1.RD2.RDN RADIUS DERIVATIVES
64
Chapter 2 Potential Problems
DIMENSION XCOH I , YCOM ) ,01 ( 4 ) ,OME( 4 )
DATA GI/0.66113631.-0.86113631,0.33998104,-0.33998H4/
DATA OME/0.34785486,0.34785485,0.65214515,0.65214515/
??.(??-??/?.
BY<Y2Yl)/2.
SLSQRT(AX*2*AY*2)
ETA1*AY/SL
ETA2-AX/SL
G>0.
H'O.
DUl=0.
DUJrO.
DQl'O.
DQ2*0.
?
? COMPUTE G, H, DO1, D92, DU1 AND DU2 COEFFITIENTS
?
DO 40 Il,4
XCOA)<AXCI(I)*BX
YCO(I)AYGI(I)*BY
RA>SQRT((XP-XCOI1))**2*(YP-YCO{J))*2)
BD2(YCO(I)-YP)/RA
RD?RD1*ETA1RD2*?TA2
1F(K) 30,30,10
10 DU)-DU1RD1OMEA)*SL/R>
DU2DU2RD2OME(I)*SL/RA
DQlsDQl-(B.RD1*2-1.)*ETA12.RD1*RD2*ETA2*OMEA)*SL/RA*2
DQ2<DO2-(B.*RD2**2-1.)*ETA2+2.RD1*RD2*ETA1)OHEA)*SL/RA**2
30 G>GALOGA/RA)*OHEA)*SL
40 HH-RDNOME(I)SL/RA
RETURN
END
Routine LOCINPC
This routine simply computes equation B.41) to obtain the diagonal elements of
G. As we have used throughout the fundamental solution ln(l/r), the formula has
to be multiplied by 2?, i.e. B.41) becomes.
1/2
SUBROUTINE LOCJNPC(X1,Y1,X2,Y2,G)
?
? PROGRAM 5
?
? THIS SUBDOi/TrNE COMPUTES THE VALUES OF THE DIAGONAL
? COEFFITIENTS OF THE G MATRIX
? I
AX(X2-Xl)/2.
AY*(Y2-Yl)/2.
SR*SQRT(AX**2+AY**2)
Gs2*SRA.-ALOG(SK))
RETURN
END
Routine SLNPD
This is a standard routine given in reference [10] which can solve the system of
equations using pivoting if needed. If the matrix A has a zero in the diagonal i
t
2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 65
will interchange rows, deciding that the system matrix is singular only when no
row interchange will produce a non-zero diagonal coefficient. If this happens it
will give a message indicating a singularity in that row.
After elimination the results are stored in the same right hand side vector DFI.
SUBROUTINE SLNPD(A,B,D,N,NX)
? PROGRAM 6
'
?
? SOLUTION OF LINEAR SYSTEMS OF EQUATIONS
? BY THE GAUSS ELIMINATION METHOD PROVIDING
? FOR INTERCHANGING ROMS WHEN ENCOUNTERING A
? ZERO DIAGONAL COEFICIENT
?
? ? : SYSTEM MATRIX
200 B(K)=B(K)-A(K,J)B(J)
66
Chapter 2 Potential Problems
? COMPUTE VALUE OF DETERMINANT
?
Dl.
DO 2S0 1*1,N
250 D*D*A(I,I)
CO TO 300
8 VRITE(,2) ?
2 FORMAT!' *** SINGULARITY IN ROW ,15)
D0.
300 RETURN
END
Routine INTERPC
Subroutine INTERPC reorders Fl (boundary condition vector) and DF1 (unknown
vector) in such a way that all the values of the potential arc stored in Fl and
all
the values of the derivatives or fluxes in DFI.
This subroutine also computes the potential values for the internal points using
formula B.32) and the fluxes along jc, and x2 directions using formula B.34).
Note that because all the ? and G terms appear multiplied by 2v the solution for
the internal points is also multiplied by 2ir.
SUBROUTINE INTEBPC(FI,DF],KODE,CX,CY,X,Y,POT,FLUX1,FLUX2I
PROGRAM 7
THIS SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL
AND THE POTENTIAL DERIVATIVES (FLUXES) AT INTERNAL POINTS
COMMON N.L.INFMPR
DIMENSION FlA),DFI(I),KODEA),CXA),CYA),XA),YA)
DIMENSION POTA),FLUX1A),FLUXZA)
REARRANGE THE Fl AND DFI ARRAYS TO STORE ALL THE VALUES OF THE
POTENTIAL IN Fl AND ALL THE VALUES OF THE DERIVATIVE IN DFI
DO 20 1*1,N
IF(KODE(D) 20,20,10
10 CHFI(I)
FKI).DFI(l)
DFI(I)CH
20 CONTINUE
COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS
IF(L.EQ.O) GO TO 50
DO 40 K'l.L
POT<K)0.
FLUX1(K)*O.
FLUX2(K)*0.
DO 30 J'l.N
KKJ+1
CALL EXTINPC(CX(K),CY(K),X(J),Y(J>,X(KK),Y(KK),A,B
I,DQ1,DQ2,DU1,DU2,1>
POT(K)POT<K)+DF1(J)*B-FIU)*A
FLVX1<K)FLUX1(K)*DFI(J)DU1-FI(J)*DQ1
30 FLUX2(K)*FLUX2(K)+DF1(J}*DU2-FI<J)*DQ2
POT(K)POT(K)/B.3.1415926)
FLUX1(X>FLUX1(K)/| 2.3.14 15926)
40 FLUX2(K)FLUX2(K)/B.3.1415926)
50 RETURN
END
Routine OUTPTPC
This routine outputs the results. It first lists the coordinates of the boundary
nodes
and the corresponding values of potential and its derivatives (or fluxes). It al
so
2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 67
prints the values of potential and fluxes at internal points if any have been
requested.
SUBROUTINE OUTPTPC(XM,YH,F1,DFI ,CX,CY,POT,F1,UX1,FLUX2)
?
? PROGRAM 8
?
? THIS SUBROUTINE PRINTS THE VALUES OF THE POTENTIAL AND ITS NORMAL
? DERIVATIVE AT BOUNDARY NODES. IT ALSO PRINTS THE VALUES OF THE
? POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
COMMON N,L,INP,IPR
DIMENSION XHU),YMA),FIA),DFIA),CXA),CYA)
DIMENSION POTA),FLUX1A),FLUX2A)
?
VRITEUPR.IOO)
100 FORMAT (' ',79( ' ' )//JX. 'RESULTS*//2X, 'BOUNDARY NODESV/8X ,'X', 15
1X,'Y",1??,'POTENTIAL',??.'POTENTIAL DERIVATIVE'/)
DO 10 1=1,N
10 VRITEUPR.200) XM(I),YN(I),FI(I),DFI(I)
ZOO FORMAT<4<2X,EH.5))
?
IF(L.EQ.O) GO TO 30
WRITE)IPR,300)
300 FORMAT<//,2X,'INTERNAL POINTS',//8X,'X',15X,'Y'.13X,'POTENTIAL',
1X,'FLUX X'.IOX.'FLUX Y'/l
DO 20 K<1,L
20 WRITE* 1PR,-OO)CX<K ),CY(K) , POT(K) ,FXUXI(K),FLUX2(K)
400 FORMAT*5BX,E]4.5)>
30 WRITE(IPR,SO0)
500 FORMAT(' ',T9('*'))
RETURN
END
Example 2.1
The following example illustrates how the code can be used to analyse a simple
potential problem. Consider the case of a square close domain of the type shown
in figure 2.8, where the boundary has been discrclizcd into 12 constant elements
with 5 internal points.
The input statements are as follows:
HEAT FLOW EXAMPLE (DATA)
HEAT FLOW EXAMPLE A2 CONSTANT ELEMENTS)
12 5
0. 0. 2. 0. 4. 0. 6. 0. 6. 2. 6. 4.
6. 6. 4. 6. 2. 6. 0. 6. 0. 4. 0. 2.
1 0
1 0
1 ?
? ?
? ?
0
?
1
?
1 ?
1 ?
? 300
? 300
? ???
2. 2. 2. 4. 3. 3. 4. 2. 4. 4.
The results arc printed out as follows.
68
Chapter 2 Potential Problems
HEAT FLOW EXAMPLE (OUTPUT)
?*********
HEAT FLOW EXAMPLE A2 CONSTANT ELEMENTS)
DATA
NUMBER OF BOUNDARY ELEMENTS 12
NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED S
COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELEMENTS
POINT X Y
?
3
4
5
?
7
8
S
10
11
12
.OOOOOEtOO
.20000Et01
4OOO0Et01
.eOOOOEtOl
,60000Et01
.60000Et01
.60000Et01
.40000Et01
.20000Et01
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
,2OOO0EtOl
.40000E+01
.60000E.01
.60000E01
,60000E*01
.60000Et01
.40O00EtOl
.2OOO0EtOl
BOUNDARY CONDITIONS
NODE
1
2
4 .
5
6
7
8
9
10
11
12
CODE
1
1
1
0
0
0
1
1
1
0
0
0
RESULTS
BOUNDARY NODES
X
.lOOOOEtOl
.3O000f>Ol
.SOOOOEtOl
.60OOOE01
.6OOOOE0l
.60000E+01
.SOOOOEtOl
.30O00E*01
.lOOOOEtOl
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
INTERNAL POINTS
X
.20000?t01
.2OOO0EtOl
.?0000?+01
.400O0EtOl
.4OOOOEtOl
PRESCRIBED VALUE
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
,3OOOOEt03
,3OO00Et03
,3OO00Et03
V
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.lOOOOEtOl
.aooooEtoi
. 60000E-101
.EOOOOEtOl
.eOOODEtOl
SOOOOEtOl
.iOOOOEtOl
,3000OEtOl
.lOOOOEtOl
Y
.ZOOOOEtOl
,40000Et01
.30OOOEtOl
.20000E01
.40000EtO)
POTENTIAL
.ZS225EtO3
. 15??2??03
.47TS0EtO2
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.47?50E*02
.15OO2EtO3
.25225EtO3
.3O000Et03
,300O0Et03
.30000Et03
POTENTIAL
.2OO28EtO3
,20028Et03
.lS001Et03
.9974OEtO2
.9974OEtO2
POTENTIAL DERIVATIVE
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.S2962EtOZ
-,4B771E02
-.S2962EtO2
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.S2969EtO2
.48737EtO2
.?2969?*02
FLUX X
-.5O3O3EtO2
-.50303?-?2
-.SO21SEtO2
-.50306Et02
-.SO3O6EtO2
FLUX Y
.14976EtOO
.1414EtOO
.4??6??-?5
. US64E00
.14S64?t00
Notice the excellent agreement of the results with the exact solution given in
figure 2.7(a)), when the coarseness of the mesh and the simplicity of the model
2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 69
q-0
= 300-*,.u=0
q= -50
7
6
?
of ? or q
M(a) Linear Element Definitions
element j+2
element j
(b) Element Intersection
Figure 2.8 Linear element. Basic definitions
The values of ? and </ at any point on the clement can be defined in terms of
their nodal values and two linear interpolation functions ?? and ??, which arc
given in terms of the homogeneous coordinate ?, as shown in figure 2.8(a),i.c.
B.45)
? is the dimensioniess coordinate varying from
I to + I and the two interpolatio
n
functions arc
A = 10 c>; V2 = ]d + c)
B.46)
Let us consider the integrals over an clement '/. Those on the left hand side
can be written as,
f m/*,/r= J [MJ
B.47)
72
Chapter 2 Potential Problems
where for each element '/* we have the two terms,
''V'~ l ?\?* '? B.48)
and
Similarly the integrals on the right hand side give
3 ??* AT j [^1^2]"* *'
where
and
B.49)
B.50)
B.51)
B.52)
Treatment of Corners
? domain discrctized using boundary elements will present a series of corners
which require special attention as the conditions on both sides may not be the s
ame.
When the boundary of the domain is discretized into linear elements, node 2
of element "j" is the same point as node 1 of element "j + ? (figure 2.8(b)). Si
nce
the potential is unique at any point of the boundary, u2 of element */ and ?1 of
element 7+ ? are both the same. However, this argument can not be applied as
a general rule to the flux, as there are boundary points for which the ftux docs
not have a unique value. This takes place at points where (he normal to the
boundary is not unique (corner points). It may also happen that the flux prescri
bed
along a smooth boundary presents discontinuities at certain particular points.
While corners with different values of the flux at both sides exist in many prac
tical
problems, discontinuous values of the flux along a smooth boundary are seldom
prescribed.
To take into account the possibility that the flux of node 2 of an element may
be different from the flux of node I of the next element, the fluxes can be arra
nged
in a 2t array.
Substituting equations B.47) and B.50) for all '/ elements into B.44) one
obtains the following equation for node T.
C""] {
4
?2
B.53)
,2,V
2.5. Linear Elements
73
where U'J is equal to the h\' term of element */' plus the hlj~x term of element
'/' ?. Hence formula B.53) represents the assembled equation for node T. Note
the simplicity of this approach. Equation B.53) can be written as,
N
2??
cV+ I /?fV = I G'V
B.54)
Similarly, as was previously sh,own (equation B.28)), this formula can be writte
n as
??
2N
?WV=IGY B.55)
and the whole set in matrix form becomes
HU = GQ B.56)
where G is now an N x 2?/ rectangular matrix.
Several situations may occur at a boundary node: First that the boundary be
smooth at the node. In such a case both fluxes 'before' and 'after' the node are
the
same unless they are prescribed as different, but in any case, only one variable
will
be unknown either the potential or the unique flux. Second, that the node is at
a
corner point. In this case four different cases are possible depending on the
boundary conditions:
(a)
Known values: fluxes 'before' and 'after' the corner.
Unknown value: potential
(b)
Known values: potential, and flux 'before' the corner.
Unknown value: flux 'after' the corner
(c)
Known values: potential, and flux 'after' the corner.
Unknown value: flux 'before" the corner
(d),Known values: potential.
Unknown values: flux 'before' and 'after' the corner.
There is only one unknown per node for the first three cases, and two unknowns
for case (d). As long as there is only one unknown per node, system B.56)can be
reordered in such a way that all the unknowns are taken to the left hand side an
d
obtain the usual system of N x N equations, i.e.
AX = F B.57)
where X is the (N) vector of unknowns; ? is the (N x N) matrix of coefficients
which columns are columns of the matrix H, and columns of the matrix G after
a change of sign or sum of two consecutive columns of G with opposite sign when
the unknown is the unique value of the flux at the corresponding node. F is the
known vector computed by the product of the known boundary conditions and
the corresponding coefficients of the G or H matrices.
74
Chapter 2 Potential Problems
When the number of unknowns at a corner node is two (case (d)), one extra
equation is needed for the node. The problem can be solved using the idea of
'discontinuous' elements [11] presented in section 2.7.
2.6 Computer Code for Potential Problems using Linear Elements
(POLINBE)
Although this code has many routines which are similar to those developed for
the constant clement case (POCONBE), iherc arc some parts which require
modification.
Main Program
The integer variables have the same meaning as in the constant elements program.
The same can be said for the real arrays except for the mid-point coordinates XM
and YM that are not needed as now the nodes are at the inlcr-clcmcnt junction.
Arrays FI and DFI now have a different meaning. The dimension of FI is {N)
while the dimension of DFI is BJV). Prescribed boundary conditions are read in
DFI (two per clement). FI is used as the right hand side vector that after solut
ion
contains the values of the unknowns. Finally both vectors are reordered to put
all the values of the potentials in FI and all the values of the fluxes in DFI a
s was
done for constant elements. Now, however FI contains one potential and DFI
two fluxes, per node.
The program allows for the flux 'before* and 'after' any node to be different.
When two, equal or different, fluxes arc prescribed at a node the potential is
computed; if the potential and one flux are prescribed, the other flux is comput
ed;
and in the case that only the potential is prescribed, both fluxes are considere
d
to be equal. It should be noticed that in problems with only one uniform region,
the case of potential prescribed and two different unknown values of the flux wi
ll
only take place in a corner where the potential is prescribed along the two elem
ents
that join at that corner. This situation is not frequent and since the potential
would be known along two different directions emerging from the corner, the
potential derivatives along these two directions would\bc known and consequently
the flux along any direction would also be known. Thus, the three variables woul
d
be known at that corner and hence any two of them can be prescribed and the
third will be computed. Notice that only for the case of a singularity on the co
rner
would one require replacing the corner node by two different nodes inside each
of the two adjacent elements.
The listing is as follows:
2.6. Computer Code for Potential Problems using Linear Elements (POLINBE)
75
?
?
PROGRAM POLINBE
?
? PROGRAM 9
?
? THIS PROGRAM SOLVES TWO DIMENSIONAL ( POTENTIAL PROBLEMS
? USING (LIN)EAR (B)OUNDARY (E)LEHENTS
?
CHARACTER*10 FILEIN.FILEOUT
COMMON/MATG/ G(80,16O)
COMMON/MATH/ ?(80,80)
COMMON N,L,INP,IPR
DIMENSION X(81),Y(B1),FI(80),DFIA60)
DIMENSION KODEA60).CX<20),CYB0),POT<20),FLUXlB0),FLUX2B0)
?
? SET HAXIMUN DIMENSION OF THE SYSTEM OF EQUATIONS (NX)
? NX = MAXIKUN NUMBER OF NODES * HAXIMUN NUMBER OF ELEMENTS
?
NX: 80
NX1=2NX
?
? ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES
?
INP=5
I PR 6
?
? READ NAMES AND OPEN FILES FOR INPUT AND OUTPUT
?
WRITE!*,' (A) ') ' NAME OF THE INPUT FILE (MAX. 10 CHART.)'
READ!*,' (A) ') FILEIN
OPEN( INP,FIl.EFILEIN,STATUS*'OLD' I
WRITE!,' (A) ) ' NAME OF THE OUTPUT FILE (MAX.10 CHART.I
REAtH*,1 (A) ') F1LEOUT
OPEN!IPR.FILE'FlLEOUT.STATUSc'NEW*)
?
? READ DATA
?
CALL INPUTPL(CX,CY,X,Y,KOD?,DFI)
?
? COMPUTE G AND ? MATRICES AND FORM SYSTEM (A X * F)
?
CALL GHKATPL(X,Y,G,H,FI,DFI,KODE,NX,NX1)
?
? SOLVE SYSTEM OF EQUATIONS
?
CALL SLNPD(H,FI,D,N,NX)
?
? COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
CALL 1NTERPLIF1,DFI,KODE,CX,CY.X.Y.POT,FLUX1.FLUX2)
?
? PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS
?
CALL OUTPTPL(X,Y,FI,DF1,CX,CY,POT,FLUX1,FLUX2)
?
CLOSE (INP)
CLOSE (IPR)
STOP
END
Routine INPUTPL
The input subroutine is similar to INPUTPC in POCONBE. Only the boundary
conditions arc prescribed in a different way. Now, two boundary conditions per
clement arc read in array DFI. Thus, each node */'* may have a different value o
f
the flux, one as the end node of clement '7 ? and the other as the start node of
clement '/"
76
Chapter 2 Potential Problem
SUBROUTINE INPUTPL(CX,CY,X,Y,KODE,DFI)
?
? PROGRAM 10
?
? N= NUMBER OF BOUNDARY ELEMENTS
? L= NUMBER OF INTERNAL POINTS
?
CHARACTER*80 TITLE
COMMON N,L,IHP,IPR
DIMENSION CXA),CYA),XA),YA),KODEA),DFIA)
WRITE(IPR,100)
100 FORMATC ',79('*'))
?
? READ JOB TITLE
?
READ(INP,'(A)') TITLE
WRITE(IPR,'(A)') TITLE
?
? READ NUMBER OF ELEMENTS AND INTERNAL POINTS
?
READ(INP,*)N,L
WRITE(IPR,300)N,L
300 FORMAT(//' DATA'//2X,'NUMBER OF BOUNDARY ELEMENTS - ,I3/2X, 'NUMBER
1
OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED =',I3)
?
? READ BOUNDARY NODES COORDINATES IN ARRAYS X AND Y
?
WRITE(IPR,500)
500 FORMAT(//2X, 'COORDINATES OF THE EXTREME POIHTS OF THE BOUNDARY ELE
1MENTS',//2X, 'POINT',10X, 'X',18X, 'Y')
READ(INP,*) <X(I),Y(I),I=1,N)
DO 10 1=1,N
10 WRITE(IPR,700I,X(I),Y(I)
700 FORMAT{3X,I3,2EX,E14.5))
?
? READ BOUNDARY CONDITIONS IH DFI(I) VECTOR.IF KODE(I)=0
? THE DFI(I) VALUE IS A KNOWN POTENTIAL; IF KODE(I)1 THE
? DFI(I) VALUE IS A KNOWN POTENTIAL DERIVATIVE (FLUX).
? TWO BOUNDARY CONDITIONS ARE READ PER ELEMENT.
? ONE NODE MAY HAVE TWO DIFFERENT VALUES OF THE
? POTENTIAL DERIVATIVE BUT ONLY ONE VALUE OF THE POTENTIAL
?
WRITE(IPR,800)
800 FORMAT(//2X,'BOUNDARY CONDITIONS'//15X, ' FIRST NODE
',
19X,' SECOND NODE
'/17X, 'PRESCRIBED', 20X, 'PRESCRIBED'/
2
, IX, 'ELEMENT* , 12X, 'VALUE' , 7X, 'CODE' , 14X, 'VALUE' , 7X, 'CODE'
)
DO 20 1=1,N
READ(INP,*) KODEB*I-1),DFIB*I~1),KODBB*I),DFIB*I)
20 WRITE(IPR,950)I,DFIBI-1),KODEB*I-1),DFIB*I),KODEB*I)
950 FORMATBX,I3,2<10X,E14.7,5X,I1))
?
? READ COORDINATES OF THE INTERNAL POINTS
?
IF(L.EQ.O) GO TO 30
READ(INP,*)
30 RETURN
END
Routine GHMATPL
Notice that this routine is similar to the one described in POCONBE with the
main difference that the y1' elements are assembled in a N x IN matrix instead o
f
an N x N one as was previously the case. This is because two possible values of
flux are considered at each node, one to the left and the other to the right of
it.
Then the boundary conditions are applied as described earlier to rearrange the
system of equations and prepare it for solving.
2.6, Computer Code for Potential Problems using Linear Elements. (POLINBE)
77
The diagonal terms in II are computed implicitly. Assuming a constant potential
over the whole boundary the (lux must be zero and hence
H! = 0 B.58)
where I is a vector that for all nodes has a unit potential. Since B.55) has to
be
satisfied
i)
B.59)
which gives the diagonal coefficients in terms of the rest of the terms of the H
matrix.
The above considerations are strictly valid for close domains. When dealing
with infinite or semi-infinite regions, equation B.56) must be modified. If a un
it
potential is prescribed for a boundless domain the integral
J ?*A? B.60)
rover the external boundary ?? at infinity will not be zero and since p* is due t
o
a unit source, this integral must be (see equation B.17) when r = e~* oo)
J ?*</?=-1
B.61)
r,
The diagonal terms for this case are,
?"=1-??? (forj^i)
B.62)
Notice that as all the terms H'J and G'J are multiplied by 2ir in the program,
because the fundamental solution has been taken as In(llr), equation B.62) is al
so
written in the program as
H" = 2w-t HiJ (for / * j)
B.63)
?
.......
SUBROUTINE GHMATPL(X,Y.G,H.F1,DF1,KODE,NX,NX1)
?
? PROGRAM 11
?
? THIS SUBROUTINE COMPUTES THE G AND H MATRICES
F
? AND FORMS THE SYSTEM OF EQUATIONS A X
? ? IS A SQUARE MATRIX (N,N>; G IS RECTANGULAR (N,2*N)
?
COMMON N.L.INP.IPR
DIMENSION X<1),YA>,C(NX,NX1),H(NX,NX),FIA),KODEA),DF!A)
NN<2*N
DO 10 Icl.N
DO 6 J.l.N
6 H(I,J)0.
DO 10 J'l.NN
10 G(I,J)0.
?
? COMPUTE THE COEFFICIENTS OF G AND H MATRICES
78
Chapter 2 Potential Problems
()
Y(Nl|*Y(l>
DO 100 1*1,N
NF'11
NS'I+N-2
DO 50 JJ*NF,NS
IF(JJ-N!30,30,20
20 J.JJ-N
CO TO 4 0
?0 3*33
40 CALL EXTINPL(X(I),Ym,X(J),Y(J),X(J-H),Y(Jl),*l,A2,Bl.B2
l>DQlFl,DQ2Fl,DQlF2lIXJF2,DUlFl,DU2Fl,i>UlF2,DU2F2,0)
1F(J-NL2,43,43
42
HU,J1)HU,J-1)+A2
GO TO 44
43
H(I.1)H(I,1)*A2
44
H(I.J)HA,J)+A1
G(It2*J-l)Bl
G<I,2*J)*B2
50 HA,I)*HU,I)-A1>A2
NF*I*N-1
NS1+N
DO 95 JJsNF.NS
IF(JJ-NO0,70,60
60 JJJ-N
GO TO 80
TO ????
80 CALL LOCINPL(X(J),Y(J),X(J)),V(J+1),B1,B2)
IF(JJ-NF)B2,82,83
82
CH'Bl
B1*B2
B2<CH
83
G(I,2*J-1)'B1
95 GA,2*J)B2
?
? AOD ??? ?? THE DIAGONAL COEFFICIENTS
? FOR EXTERNAL PROBLEMS.
?
IF(H(I,1>> 98,100,100
98 H(I,I)-6.2831852H(I,I)
100 CONTINUE
?
? REORDER THE COLUMNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE
? WITH THE BOUNDARY CONDITIONS AND FORM THE SYSTEM MATRIX A
? WHICH IS STORED IN H
?
DO 155 I1,N
DO 150 J'1,2
IF(KODE<2*J-2*J) MIO.l?, 150
110 IFII.NE.N .OR. J.NE.2) GO TO 125
IF(KODEU)) 115,115,113
113
DO 114 K1,N
CH'H(X,1)
H(K,1)--G(K,2N)
114
G(K,2*N)-CH
GO TO ISO
115
DO 116 K*1,N
K(K,1)H(K,1)-C(K,2*N)
IK G(X,2*N)'O.
CO TO 150
12S IFU.EO-1 .OR. J.GT.l .OR. XODEf 2*1-2).EQ. 1) GO TO 130
DO 129 K*1,N
H(K,I)H(X,I)-G(K,2*I-1)
129
G<K,2*1-11*0.
GO TO 150
130
DO 132 K*1,N
CHH(X,I-1J)
H(X,I-1J)-G(K,2*I-2*J)
132 G(K,2*I-2*J1*-CH
150 CONTINUE
155 CONTINUE
?
? FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN FI
?
DO 160 1*1,N
DO 160 J1,NN
FI(I)*FI<I}*G(I,J)*DFI(J)
160 CONTINUE
RETURN
END
2.6. Computer Code for Potential Problems using Linear Elements (POLINBE)
79
Routine EXTINPL
This routine is similar to the one in POCONBE but instead of computing only
one value per element for each coefficient as in POCONBE, it now computes
two values per element, i.e. the parts of the coefficients corresponding to the
adjacent nodes,
?
SUBROUTINE EXTINPLIXP.YP.Xl,V1,X2,Y2,A1,A2,B1,B2
1.DQ1F1,DQ2F1 ,D<J1F2 ,D<S2F2,DU1F1 ,DU2F1 ,DU1F2,DU2F2,K)
?
?
PROGRAM 12
?
?
THIS SUBROUTINE CONPUTES THE INTEGRAL OF SEVERAL NON-SINGULAR
?
FUNCTIONS ALONG THE BOUNDARY ELEMENTS USING A FOUR POINTS
?
GAUSS QUADRATURE
?
WHEN K0, THE OFF DIAGONAL COEFFICIENTS OF THE ? AND ? MATRICES
?
ARE COMPUTED
?
WHEN K=l, ALL THE COEFFICIENTS NEEDED FOR COMPUTATION
?
OF THE POTENTIAL AND FLUXES AT INTERNAL POINTS ARE
?
COMPUTED (A1,A2.B1,B2,FU,F21,F12,F22,E11,E21,E12,E22}
?
DISTANCE FROM THE COLOCATION POINT TO THE
?
RA RADIUS
?
GAUSS INTEGRATION POINTS ON THE BOUNDARY ELEMENTS
?
???1.???2 * COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
?
RD),RD2,RDN
RADIUS DERIVATIVES
?
DIMENSION XCOD),YCOD),G]D>,OMED>
DATA fil/0.861 13631,-0.861136.11 ,0.33998104,-0.33998104/
DATA OME/0.34785485,0.34785485,0.65214515,0.65214615/
?
????2-??/2
BX*(X2*Xl>/2
AY(Y2-Yl)/2
BY*(Y2*Yl)/2
SLeSQRT(AX*2*AY**2)
ETA1 AY/SL
ETA2'-AX/SL
Al'O.
A2O.
BJO.
B2-0.
DQ1F1*O.
DQ2F1-0.
DQ1F2O.
DO2F2=O.
DUIFI'O.
DU2F1O.
DU1F2=O.
DU2F2*0.
?
? COMPUTE THE TERMS TO BE INCLUDED IN THE G AND H MATRICES
? OR THE TERMS NEEDED FOR COMPUTATION OF THE POTENTIAL
? AND FLUXES AT INTERNAL POINTS
?
DO 40 1-1,4
XCO(I)*AX*GI(I)*BX
YCO(I)AY*GIA)*BY
RASQRT((XP-XCO(I))24(YP-YCO{I))**2)
RD1>(XCO(I)-XP)/RA
RD2(YCO(I)-YP)/RA
RDN*RD3*ETAJRD2*ETA2
IF(K) 30,30,10
10 DU1>RD1*OME(I)*SL/RA
DU2RD2*OME(I)*SL/RA
Dgis-(B.RDI**2-1.)*ETAH2.RD1RD2ETA2)OME(I)SL/RA*2
DJ2 = -(<2.*RD22-1.)ETA242.RD1RD2*ETA1)OME(ISL/RA*2
M1F1'D91F1DQ1*F1
DQlF2*DelF2*DQl*F2
DQ2FlsD2Fl4DQ2*Fl
DQ2F2>DQ2F2tDQ2*F2
DU1F1-DUIF1*DU1F1
DU1F2DU1F2DU1*F2
DUZF1*DU2F1*DU2*F1
DU2F2DU2F2*DU2F2
80
Chapter 2 Potential Problems
30 H*RDN*OME(I)*Sl/RA
G'ALOGI1/RA)*OHE(])*SL
A1*A1-F1*H
A2=A2-F2H
B]*B1F1*C
40 B2*B2F2*G
RETURN
END
Routine LOC1NPL
This routine computes now the part of the elements of the matrix G corresponding
to the integrals along the elements which include the singularity. These integra
ls
are:
?\ =
B.64)
?2 = J ?2 ? (- J </?
Using the local system of coordinates in figure 2.9, the integrals can be writte
n as
dr,
Point B)
/.\
1
/ .
B\= J A -n)ln [-) dr = I $(l - r,) In l~
B2=
Poim B)
Poinl A)
B.65)
B.66)
Figure 2.9 Linear element local coordinate system
2.6. Computer Code for Potential Problems using Linear Elements (POL1NBE)
81
SUBROUTINE LOCINPL(X>,Y1.X2,Y2,B1.B2)
?
? PROGRAM IS
?
? THIS SUBROUTINE COMPUTES THE PARTS OF THE ? MATRIX
? COEFFICIENTS CORRESPONDING TO INTEGRALS ALONG AN ELEMENT
? THAT INCLUDES THE COLLOCATION POINT.
?
SEPSQRT(IXZ-X1)**2*(Y2-Y1)Z)
B1-SEP4 1.5-AI.OC(SEP))/2
B2SEP(O.5-ALOG(SEI?))/2
RETURN
END
Routine INTERPL
This routine replaces the INTER PC program used in POCONBE. Il first arranges
all potentials in FI and all their derivatives (or fluxes) in DFI and then compu
tes
the values of potentials and fluxes at the internal points if requested.
SUBROUTINE INTERPL(Fl,DFl,KODE,CX,CY,X,Y,POT,FtUXl,FLUX2)
?
? PROGRAM 14
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL
? AND THE POTENTIAL DERIVATIVES (FLUXES) AT INTERNAL POINTS
?
COMMON N.L.INP.IPR
DIMENSION Fll1),DFIA),KODE(I),CXA I,CYA),XA>,Y{1)
DIMENSION POT1 ) ) .FLUX1( 1).FLUX2 A )
?
? REARRANGE THE FI AND DFI ARRAYS TO STORE ALL THE VALUES OF THE
? POTENTIAL IN FI AND ALL THE VALUES OF THE DERIVATIVE IN DFI
?
DO 155 1*1,N
DO 150 J*l,2
IF(KODEB*I-2*JI)I10,110,150
110 JKII.NE.N .OR. J.NE.2) CO TO 125
IKKODE(l)) 1L,114,113
113
CH*FI<1)
FHUDFI<2N)
DF1B>N)*CH
CO TO 150
114
DFlB*N)sDF!(l)
GO TO 150
125 IFU.EQ.l .OR. J.EQ.2 .OR. KODEB*1-2).EC.? GO TO 130
DFlBl-l)=DFK2I-2)
CO TO J50
130 CH*FIU-l*J)
DF1B*1-24J)CH
150 CONTINUE
155 CONTINUE
?
? COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
IF(L.EQ.O) GO TO 50
DO 40 K=J,L
POT(K)*0.
FLUX1(K)O.
FLUX2(K)0.
DO 30 J1,N
CALL EXTlNPL{CX(K),CY(K),X(J),r(J).X(J*l),Y(Jtl),Al,A2,Bl,B2
1,DQ1F1,DQ2F1,DQ1F2,DQ.2F2,DU1F1,I>U2F1.DU1F2,DU2F2,1)
1F(J-NK2,33,33
S2 POT<K)-POT(K)PFIB*J-1)B1DFIBJ)B2-F1(J)A1-FI(J1)A2
FLUX1(K)<FLUX1(K)'DFIB*J-1)>DU1F1'DFIB*J)*DU1F2
1-FI(J)*DQ1F1-FI(Jl)*DQ1FZ
FLUX2(K)>FLUX2<K)*DFIBJ-l)*DU2FltDFlB*J)*DU2F2
1-F1(J)*M2F1-FI(J>H*M2F2
00 TO 30
82
Chapter 2 Potential Problems
33 POT(K)*FOT(R)DFIB*J-l)*BWDFlBJ)*B2-FI(j)*Al-FI(l)*A2
FLUXMK>FUntl(K)*DPIB*J-l)*I>UlFl'DFl<2J)*DUlF2
J-FI(J)DeiFl-FI(l)D<aiF2
rLUX2(K)FLUX2<K)*DFlB*.l-l)*DU2Fl4DFIB*J)*DU2F2
10
11
12
-I+
6
2003 99.7
"V-50.3) -*?-50.3)
1500
+-50.3
2003
997
(-50.3) (-50.3)
-52.9
-48.7
-52.9
1 * ]
I 47 8
150 0
2522
(a) Results obtained with
the constant element code
(P0C0NBE) 12 elements solution
10
11
12'
,2000 ,100.0
"?- 50.0) - 50.0)
-H
1500
(-50.0)
*200.0 4000
(-50.0) (-50.01
-50.0
-50.0
-SO.0
1-50.0
1000
2000
3000
(b) Results obtained with
the linear element code
(POLINBE) 12 elements solution
Figure 2.10 Results obtained using constant and linear elements for the heat flo
w
example
84
Chapter 2 Potential Problems
4?014 _L100
M-52.2) M-49.2)
,149.8
-500
300
Figure 2.10 continued
(c) Results obtained with
the linear element code
(POLINBE) 4 elements solution
The data corresponding to the 12 element problem is as follows.
HEAT FLOW EXAMPLE (DATA)
HEAT FLOW EXAMPLE A2 LINEAR ELEMENTS)
12 5
0. 0. 2. 0. 4. 0. 6. 0. 6. 2. 6. 4. 6. 6.
6. 2. 6. 0. 6. 0. 4. 0. 2.
1 0.
1 0.
1 0.
0 0.
0 0.
0
0.
1
0.
1 0.
1 0.
0 300.
0 300.
0 300.
4.
1 0.
1 0.
1 0.
0 0.
? ?.
0
?.
1
0. 1
1 0. 1
1 0. 1
0 300.
0 300.
0 300.
2. 2. 2. 4. 3. 3. 4. 2. 4. 4.
and the output is given by
HEAT FLOW EXAMPLE (OUTPUT)
HEAT FLOW EXAMPLE A2 LINEAR ELEMENTS)
DATA
HVMBER OF BOUNDARY ELEMENTS < 12
NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED
2.6. Computer Code for Potential Problems using Linear Elements (POL1NBE)
85
COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELEMENTS
POINT
1
2
3
4
S
6
7
8
9
10
11
12
X
.OOOOOEtOO
.20000Et01
.<0000?*0l
,60000Et01
.60000Et01
.OOOOEtOl
.60000??]
.4OOOOEtOl
.zooooEtoi
.OOOOOEtOO
.000001*00
.OOOOOEtOO
Y.
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.ZOOOOEtOl
.40000E01
.60000E401
.6000QEt01
.SOOOOEtOl
.60000Et0]
.4OO00EtOl
'. .20000EtOl
BOUNDARY CONDITIONS
ELEMENT
1
2
3
4
?
6
T
8
9
10
?
12
RESULTS
BOUNDARY
FIRST NODEPRESCRIBED
VALUE
.OOOOOOOEtOO
.OOOOOOOEtOO
.0O000OOE+0O
,OO00O00EOO
.OOOOOOOEtOO
.000O000EO0
.000O0O0E00
.OOOOOOOEtOO
.OOOOOOOEtOO
. 30OOO0OE-.O3
.3OOO0OOE03
,3O0OOOOE*O3
NODES
CODE
1
I
]
0
0
0
1
1
1
0
0
0
SECOND NODEPRESCRIBED
VALUE
. 0000000??0
.0000000?00
.OOOOOOOEtOO
.0000000?00
,0000OOOE0O
.OOOO0O0E*00
.OOOOOO0E*OO
,000OOO0E0O
.OOOOOOOE400
.3000000E*03
.300O0OOE+03
.30OOOO0EO3
CODE
1
1
1
0
0
0
1
1
]
0
0
0
.00000E00
.20O0OEOl
.40000E+01
.60000E01
.60000E401
.60000E0)
,60000E01
.40000E01
.200001401
.OOOOOEtOO
.OO000E00
.O00O0E*0O
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.ZOOOOEtOl
.40000Et01
.60000Et01
.60000EtOI
.eOOOOEtO1
.60000fct01
.4O000Et01
.JOOOOEtOl
POTENTIAL
.3OOOOE+O3
.2OOOOEtO3
.10000?40?
.OOOOOEtOO
.OOOOOEtOO
,00000E*00
.OOOOOEtOO
.2OO00Et03
.3OO0OEt03
.300O0Et03
.30?00??03
POTENTIAL DERIVATIVE
BEFORE NODE AFTER NODE
.S0O00Et02
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.S00O0EtO2
.&OOOOEtO2
.SOOOOEtOJ
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.@O00EtO2
.6OOOOE02
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
-.S0O0OEt02
-,SOOOOEtO2
-,SOOOOEtO2
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.SOOOOEtOS
.S0OO0EO2
.6OO00Et02
INTERNAL POINTS
X
.20000E01
.20O0OE01
.30000E401
.40OOOE.01
.4OOOOEtOl
.20000E0)
.40000Et01
3OOOO?tOJ
. 20000??]
.40000t.01
POTENTIAL
.20000E03
.Z0000Et03
.lS000Et03
.10000Et03
.10000Et03
FLUX X
.50001Et02
.50001Et02
.SOOO0EtO2
.sooooEtoa
.SOOOOEtO2
FLUX Y
-.69996E-03
.6S619E-03
-.99891E-06
-.346I8E-03
,34851E-O3
****************** * ** * * *
Kxampk' 2.3
It is interesting to note that in this case even a simple four elements represen
tation
can give exact results using double value of the flux at the corners (figure 2.I
O(c)).
86
Chapter 2 Potential Problems
The input in ihis case is
HEAT FLOW EXAMPLE (DATA)
HEAT FLOW EXAMPLE D LINEAR ELEMENTS)
4 3
0. 0. 6. 0. 6. 6. 0. 6.
1 0. 1 0.
0
0. 0 0.
1
0. 1 0.
0 300. 0 300.
2. 4. 3. 3. 4. 4.
The corresponding output is very accurate taking into consideration the
simplicity of the mesh.
HEAT FLOW EXAMPLE (OUTPUT)
HEAT FLOW EXAMPLE D LINEAR ELEMENTS)
NUMBER OF BOUNDARY ELEMENTS 4
NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED
COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELEMENTS
POINT
1
2
3
4
X
.00000E+00
.60000E401
,60000E01
.OOOOOE+00
Y
.OOOOOEtOO
. ??0???-??
.60O00E*01
.60000E01
BOUNDARY CONDITIONS
FIRST NODE
SECOND NODE
PRESCRIBED
PRESCRIBED
ELEMENT VALUE CODE
VALUE CODE
1
.OOOODOOE+00 1 .OOOOOOOE400 i
2
.00O000OE+00 0 .OOOOOOOE+QO 0
3
. OOOOOOOE+00 1 .0000000?-?0 1
4
.3OOOOOOEO3 0 . ?000000?-?? 0
?******************************************************************************
RESULTS
BOUNDARY NODES
POTENTIAL DERIVATIVE
X Y
POTENTIAL BEFORE NODE AFTER NODE
.OOOOOEfOO .OOOOOE+00 .????0?+?? .5OOOOEO2 .OOO0OE00
.60000F01 .000???-?0 .0O00OE+O0 . OOOOOE*00 -.5??0???2
.??000??1 .60000E*01 .OO0OOE00 -.50flO0F*02 .OOOOOE400
.0O0OOE00 .60000E+01 .3OOOOEO3 .00000t<00 .S0000E02
INTERNAL POINTS
X Y
POTENTIAL FLUX X FLUX Y
.20000E401 .40000E+OI .20138E03 -.52208E*02 .26160E01
.
The coefficient c' is equal to 0.5 for the nodes on discontinuous elements. The
integrals h<j, ?, gf, and g% along the discontinuous elements given by equation
B.72) can be computed by the usual Gaussian numerical quadrature when the
node " does not belong to the element. When *'/" is one of the nodes of the
discontinuous element h'J = Ajf = 0 and g'J and g'? can be easily obtained by
analytical integration. The element is subdivided into two parts one at each sid
e
2.7. Quadratic and Higher Order Elements
89
of the node. The resulting integrals consist of the same basic integrals as thos
e
of the regular linear elements given by equation B.66).
2.8 Quadratic and Higher. Order Elements
It is usually more convenient for arbitrary geometries to implement some type of
curvilinear elements. The simplest of these are the three noded quadratic elemen
ts
which require working with transformations. Consider the curved boundary shown
in figure 2.12 where ? is defined along the boundary and the r position vector i
s
a function of the cartesian system, .x,, .v2. The variables ? or q can be writte
n in
terms of interpolation functions which are functions of the homogeneous coordina
te
Z. i.e.
= ?? + ?2?2
</(C) = ?1?1 + ?2?2 + ???* = 1?1?2?3]
where the interpolation functions are
</*, = \? - i); ?? - (i - i
B.73)
B.74)
*2
'-'(1. 2)
Figure 2.12 Curved boundary
90
Chapter 2 Potential Problems
u or q variation
Figure 2.13 Quadratic element
A)
B)
1/2
112
Refetence system
These functions are quadratic in ? and give the nodal values of the variable ? o
r
</ when specialized for the nodes; i.e. with reference to figure 2.13
B.75)
Node
1
2
3
-1
0
+ 1
??
1
0
0
??
0
1
0
??
0
0
1
The integrals along any '/'element are similar to those computed for the linear
element, but there are now three nodal unknowns and the integration requires
the use of a Jacobian. Consider for instance an integral for the H type terms, i
.e.
where
B.76)
B.77)
2.8. Quadratic and Higher Order Elements
91
The evaluation ?? these terms requires the use of a Jacobian as the ?/ functions
are expressed in terms of ?, but the integrals are functions of ?. For a curve s
uch
as given in figure 2.12, the transformation is simple,
where \G\ is the Jacobian. Hence one can write,
Node 2
l
J
||
B.79)
? ?
Node 1
Formulae such as B.66) are generally too difficult to integrate analytically and
numerical integration must be used in all cases, including those elements with a
singularity.
Notice that in order to calculate the value of the Jacobian \G\ in B.78) one
needs to know the variation of the .v, and .v2 coordinates in terms of f. This c
an
be done by defining the geometrical shape of the element in the same way as the
variables ? and q are defined, i.e. using quadratic interpolation,
( )
x2 = ?2?\ + ?2?22 + ?2?1
where the superscript indicates the number of the node. This is a similar concep
t
to the isoparametric elements commonly used in finite elements.
Cubic Elements
Elements of order higher than quadratic are seldom used in practice, but they
may be interesting in some particular applications. Because of this we will brie
fly
describe the case of elements with cubic variation of geometry, and ? or q varia
bles.
In this case the functions are described by taking four nodes over each element
(figure 2.14).
?2?2 +??\?* + ???*
q = ?1 ?2 ?* ?4
and similarly
v, =* ?,?| + ?2?\
?2 = ?\ + ?\ + ?$ + ?*
92
--1
{--1/3
(?
C)
D)
Chapter 2 Potential Problems
* Xj
B)
Reference element
Figure 2.14 Cubic elements with Tour nodes
where the interpolation functions are,
which can be specialized at the nodes as follows,
-10
B.83)
Node
1
2
3
4
z
-1
-1/3
1/3
1
??
1
0
0
0
4>2
0
1
0
0
?
0
0
1
0
??
0
0
0
1
B.84)
2.8. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 93
Another possibility with cubic elements is to define the variation of ? or q in
terms of the function and its derivative along (he element, at the two end point
s
- i.e nodes I and 2 - as shown in figure 2.15. The corresponding function for ?
(same applies for q and xlx2) is then given by
21-~
with
B.85)
B.86)
where / is the element length.
X2
Node 1
Node 2
-//2-//2Reference element
Kigurc 2.15 Cubic elements with only two nodes
94
Chapter 2 Potential Problems
This last lypc of cubic clement could be used in cases where we wish lo have
a correct definition of the derivative along ?, for instance to calculate fluxes
in
that direction, or if we prefer to reduce the number of nodes along the element,
in some cases it may still be better to continue defining the geometry with four
nodes as it is generally more difficult (o have accurate results for the slopes.
2.9 Computer Code for Potential Problems using Quadratic Elements
(POQUABE)
In what follows a FORTRAN code for potential problems using quadratic elements
is described. The program has the same organization as the two previously studie
d.
All variables in the code have the same meaning as for the linear element
program (POLINBE). FI and DFI have a slightly different form. The dimension of
FI is (N), N being the number of nodes and that of DFI is ONE), where NE is
the number of elements. The prescribed boundary conditions are read in DFI
(three per element).
The program allows for the values of flux at both sides of the nodes connecting
two elements to be different. Then, (i) when both fluxes are prescribed as diffe
rent
at both sides of the node, the potential is the only unknown; (ii) when the pote
ntial
and one flux arc prescribed, the other flux is the unknown and (iii) if only the
potential is prescribed, one value of the (lux is the unknown and will be the sa
me
on both sides of the node. Thus, the situation at corner nodes of quadratic elem
ents
is the same as for linear elements.
Main Program
The program follows the same structure of the constant and linear potential code
s.
The listing is as follows.
PROGRAM POQUABE
?
? PROGRAM 16
?
?
? THIS PROGRAM SOLVES TWO DIMENSIONAL (PO)TENTIAL PROBLEMS
? USING (QUA)MMTIC (B)OUNDARY (E)LEHENTS.
?
CHARACTER10 PILEIN.FILEOUT
COMHON/HATG/GA00,150)
COKMON/MATH/HA00,100)
COMMON N.L.INP^PR
DIMENSION XA01),YA01)
DIMENSION DFIU50) ,FI 1100) ,K0DE( 150)
DIMENSION CXB0).CYB0),POTB0),FLUX1BO),FLUX2BO)
?
? SET MAXIMUM DIMENSION OF THE SYSTEM OF EQUATIONS (NX)
? NX' MAXIMUN NUMBER OF NODES' 2MAXIMUM NUMBER OF ELEMENTS
? NX1 3MAXIMUN NUMBER ELEMENTS
?
NX100
NXl'lSO
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 95
?
? ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES
?
INP5
IPR6
?
y
printed in the subroutine OUTPTPQ.
The FORTRAN listing of 1NPUTPQ is as follows:
SUBROUTINE INPUTPe(CX,CY,X,Y,ODE,DFI)
?
? PROGRAM 17
?
? ?? NUMBER OF BOUNDARY ELEMENTS
? N
NUMBER OF BOUNDARY NODES 2 * NE
? L HUHBER OF INTERNAL POINTS
?
CHARACTER'S!} TITLE
COMMON N.L.INP.IPR
DIMENSION KonE(l),X(l),T<l>.CX<l),CY(t),DFI(l)
WRITE(IPR.IOO)
100 FORMAT! ' ',7!''))
?
? READ JOB TITLE
?
READUNP,'(A)'1 TITLE
WRITEdPR,'(A)') TITLE
?
? READ NUMBER OF BOUNDARY ELEMENTS AND INTERNAL POINTS
?
READ(INP,*)NE,L
WRITE!IPR,210)NE,L
210 FORHATi//2X,'DATA'/2X, 'NUMBER OF BOUNDARY ELEHENTS*',
1I3/2X,'NUMBER OF INTERNAL POINTS',13)
N2*NE
?
? READ BOUNDARY NODES COORDINATES IN ARRAYS X AND Y
?
WRITE!IPR,500)
READUMP,*) <X(I),YA),I1,N|
DO 10 I'l.N
10 WRITEdPR,240) I,XII),Y(I)
300 FORMAT!//2X,'BOUNDARY NODES COORDINATES'///,
l'NODE'.IOX.'X'.ieX.'Y1/)
240 FORHAT(SX,13.2EX,EH.7)>
?
? READ BOUNDARY CONDITIONS IN DFId) VECTOR, IF KODEd)<0
? THE DFI!I) VALUE IS A KNOWN POTENTIAL; IF XODEd)*] THE
? DFId) VALUE IS A KNOWN POTENTIAL DERIVATIVE (FLUX).
? THREE BOUNDARY CONDITIONS ARE READ PER ELEMENT.
? NODES BETWEEN TWO ELEMENTS MAY HAVE TWO DIFFERENT VALUES
? OF THE POTENTIAL DERIVATIVE BUT ONLY ONE VALUE OF THE
? POTENTIAL.
?
WRITEdPR, BOO)
800 FORMAT!//2X,'BOUNDARY CONDITIONSV/lIX,
1
FIRST NODE -',3X,'
SECOND NODE
",3X,
j><
THIRD NODE
V13X, 'PRESCRIBED', 14X,
3 PRESCRIBED',MX,'PRESCRIBED'/IX,'ELEMENT'.X,'VALUE',
47X,'CODE',8X.'VALUE',7X,'CODE',SX,'VALUE',7X,'CODE')
DO 20 I'l.NE
READ!INF,) (KODE( JI-34j),DFH3I-3*J).J'l .3)
20 VRlTEdPR,95O)l,IDFICI-3J),KODECl-3*JKJl,3)
50 FORMAT<3X,I3,2X,3DX,E14.7,SX,I1))
?
? READ COORDINATES OF THE INTERNAL POINTS
?
XF(L.EQ.O) 00 TO 30
READdNP,*) (CXd),CY<I),Il,L)
30 RETURN
END
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 97
Routine GHMATPQ
This subroutine computes the G and H system matrices by calling routines
EXTINPQ and LOCINPQ.
EXTINPQ: Computes the GW and HW C) submatrices which relate a
collocation point with an clement as defined by its three nodes.
The collocation point is not any one of the element nodes.
LOCINPQ: Computes the GW C) submatrix for the case when the collocation
point is one of the nodes within the element under consideration
(i.e. the singularity is in the same clement). Notice that the correcorresponding HW C) is computed using EXTINPQ because the
singularity will occur only on the diagonal term and this is
computed later on by adding the oil-diagonal terms of the row.
The resulting GW and HW submatrices arc assembled in the G and H system
matrices. Matrix G is now rectangular since each extreme node of an element may
have different fluxes, i.e. one 'before' and another 'after' the node. The diago
nal
terms in H arc computed using constant potential considerations, which results
in adding row coefficients together.
Once the matrices H and G arc assembled, the system of equations needs to
be reordered in accordance with the boundary conditions to form
AX = F
where X is a [N) vector of unknowns, N being the number of nodes; A is a (N x N)
matrix whose columns arc a combination of columns of H or G depending on
the boundary conditions or of two consecutive columns of G when the unknown is
the unique value of the tractions at both sides of the extreme node of an elemen
t;
F is a known vector computed by multiplying the prescribed boundary conditions
by the corresponding row terms of G or H.
At the end of the subroutine GHMATPQ and after rearranging H contains
the matrix A, and FI the F vector.
The FORTRAN listing of GHMATPQ is as follows.
SUBROUTINE GHHATPq(X,Y.G,H,FI,DFl.RODE,NX,NXl)
?
? PROGRAM IS
?
? ThIS SUBROUTINE COMPUTES THE ? AND H MATRICES AND FORMS
? THE SYSTEM OF EQUATIONS A X
F
? ? IS A SQUARE MATRIX <Z*NE,2*NE); ? IS RECTANGULAR B*NE,3NE)
?
DIMENSION XA),YA),G(NX,NXI),H(NX,NX)
DIMENSION HK<3),GVC) ,Dgivn),DQ2VC),DU14C).DU2WC|
DIMENSION FIA>,DF1A),KODEA)
COMMON N.ImINP.IPR
NEN/2
DO 20 Il,N
DO 11 J1,N
11 HA,J)=O.
DO 12 J1,3*NE
98
Chapter 2 Potential Problems
12 G(I,J)0.
20
CONTINUE
X<N1)*XA>
Y(N*1)YA)
?
? COMPUTE THE GW AND HW MATRICES FOR EACH COLLOCATION
? POINT AND EACH BOUNDARY ELEMENT
?
DO 40 LL*1,N
DO 40 I*1,N-1,2
IF<(i,L-l>*<LL-I-l)*(LL-I-2)*(LL-I*N-2)) 22,21,22
21
NODO*LL-I1
IF( (LL.Eg.l LAND. (I.Eg.N-1)) NODO>NODON
CALL EXTINPQ(X(LL)lY(LL),X<l),V(l),X|Il),Y(I'>l),X(It2),Y(I2)
,HW,GV,DO1W,DQ2W,DU1W,DU2W,O)
CALL LOC1NPQ(XA),Y(I),X()-1),Y(I1>,X(I2) ,Y(I2),GV,NODO>
GO TO 34
22
CALL EXTINP4(X(LL),Y(LL),XU>.Y(i),X(I + l),Y<Il),X(l*2),Y(I*2)
*,HW,GW,DlW,D2W,DUtW,DU2U,0)
?
? PLUG THE CU AND HW MATRICES INTO THE GENERAL G AND H MATRICES.
?
34
DO ?? J],3
Kc3*(I-i>/2
G(LL,K-J)'C{LL,K*J)GW( J)
IF(I-N1) 37,35,37
35
IFU-3) 37,36,36
36
HILL,1)H(LL,1)HW(J)
GO TO 38
37
H(LL,I-IJ)H(LL,I-1J)HW(J)
3t CONTINUE
40 CONTINUE
?
? COMPUTE ??? DIAGONAL COEFFICIENTS OF THE H MATRIX
?
DO 70 I1,N
H<1,I)*O.
DO 60 J1,N
IF(I.EQ.J) CO TO 60
H(l,1)H(I,1)-H(I,J|
60 CONTINUE
?
? ADD ONE TO THE DIAOONAL COEFFICIENTS FOR
? EXTERNAL PROBLEMS.
?
IF(H(I,I)I 65,70.70
65 HA,I)'6.28318S2*H(I,I)
70 CONTINUE
?
? REORDER THE COLUMNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE
? WITH THE BOUNDARY CONDITIONS AND FORM SYSTEM MATRIX A WHICH
? IS STORED IN H
?
DO 180 I'l.NE
DO 170 Jl,3
IF(KODEC*I-3J)) 110,110,170
110 IFUI-NEI.NE.O .OR. J.NE.3) GO TO 12S
IFIKODE(D) 115,115,113
113
DO 114 K'l.N
CH=H(K,1)
H(K,1}*-G<K,3*I)
114
G(K,3J)>-CH
GO TO 170
115
DO 116 K'l.N
H(K,1)H(K,1)-G(K,3*I)
116
C(X,3>!)'0.
CO TO 170
12S IFI1.EQ.1 .OR. J.CT.l .OR. KODEC*1-3).EQ.1) GO TO 130
DO 129 K1,N
H(X, 2*1-D'HIK,2*1-1 )-G(K, 3*1-2)
129
G(X,3*1-21-0.
GO TO 170
130
DO 132 K>1,N
CHsH(K,2*I-2*J)
H(K,2*1-2+J)>-G(K,3*1-3*t)
132 G(K,3*I-3J)i-CH
170 CONTINUE
180 CONTINUE
?
? FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN FI
?
DO 190 1*1,N
Fill)* 0.
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 99
DO 18S J1,3NE
18S FI(I)*FIALG(I,J)*DFI(J)
190 CONTINUE
RETURN
END
Routine EXTINPQ
This subroutine computes using numerical integration, the C) submatrices GVV
and HVV that correspond to an element when the collocation point is at a node
other than any of those 3 in the element. The correlation of the collocation poi
nts
are XP and YP. The integrals are of the type B.76), i.e.
HW = J fa* cir = J ^q*|G| di B.87)
r,
-1
GW = J ??* </? = J ^u*|G| </S B.88)
This subroutine also computes the C) submatrices DU1W, DU2W, DQ1W and
DQ2W which are needed to obtain the jr, and x2 fluxes at internal points
DVIW = $ ?[~-\ itr
B.89)
DU2W = f 0(^J </?
B.90)
B.91)
The Jacobians are calculated by taking derivatives of the expressions for the .X
i
and v, coordinates, which are defined as follows (equation B.80)),
B.93)
After substituting the 0, expression (equation B.74)) the above relationships ca
n
be written as,
v, = 2i2(x! -2x\ + x\) + ?*\~x\) + x2
B 94)
2
100
Chapter 2 Potential Problems
The Jacobian is obtained by substituting B.94) into B.78) which gives
+ i2(xl-xl2J}2Va
B-95)
A Gauss quadrature formula with ten points has been taken instead of the
four points formula applied in the constant and linear clement cases. The reason
is twofold: (i) The variation of potential and flux is quadratic and hence more
points should be taken and (ii) the clement geometry is also quadratic.
SUBROUTINE EXTlNPq<XP,YP,Xl.Yl,X2,Y2.X3,Y3.HV,GW
1 ,DQ1W,DQ2V,DU1H,DU2W,IO
?
? PROGRAM 19
?
? THIS SUBRUOTINE COMPUTES THE HW AND GV MATRICES
? WHICH RELATE A NODE (XP.YP) WITH A BOUNDARY
? ELEMENT USING GAUSS QUADRATURE
? IT ALSO COMPUTES (WHEN Kl) THE D01W, DQ2V, DU1W AND DU2V MATRICES
? WHICH RELATE AN INTERNAL POINT WITH A BOUNDARY ELEMENT AND ARE
? NEEDED FOR COMPUTATION OF THE INTERNAL FLUX VALUES
?
?
RA
?
RD1.RD2.RDN
?
ETA1.ETAZ
?
XCO.YCO
?
XJA
RADIUS
RADIUS DERIVATIVES
COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
INTEGRATION POINT ALONG THE ELEMENT
JACOBIAN
?
COMMON N.L.INP.IPR
DIMENSION FC),GWC),HVC),DQ1WC),DQ2WC),DU1VC),DU2WC)
DIMENSION GIA0),OHE<>0)
DATA GI/0.9739065285,-0.973*065285,0.8650833666,-0.8650633666
,0.67940956??,-0.794095682,?.4333953941,-?.433?95341,
0.1488743389,-0.1486743389/
DATA OME/0.0666713443,0.0666713443,0.1494SI349],0.1494513491
,0.2190863625,0.2190863625,0.2692667193,0.2692667193,
0.2955242247,0.2955242247/
90 10 Jl,3
<3W<J)0.
txaiV(J)O.
D42W(J).0.
DU1V(J).O.
10 DU2W(J).O.
A-X3-2X2+X1
B(X3-X!)/2
CY3-2Y2Y1
D{Y3-Yl)/2
DO 40 Il,10
?
? COMPUTE THE VALUES OF THE SHAPE FUNCTIONS AT THE
? INTEGRATION POINTS
?
FB)I.-GI(I)**2
?
? COMPUTE GEOMETRICAL PROPERTIES AT THE INTEGRATION POINTS
?
XCOX1*FA)-X2FB)X3*FC)
YCOY1F<1)+Y2*FB>*Y3F<3)
XJASQRT((GI(I)*AB)**2(C1(I)*C*D)**2)
ETAU(GI(I)C*D)/XJA
ETA2-(GKI)AtB)/XJA
RAS4RT((XP-XCO(2*(YP-YCO)*2)
RD1'(XCO-XP)/RA
RD2(YCO-YP)/RA
RDNRD1*ETA1RD2ETA2
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 101
?
second part ot the variable n = Z- Now each singular part of these two integrals
is computed using the special integration formula, and the two non-singular part
s
together are integrated using standard 10 points Gauss quadrature.
In the program XJA1 and XJA11 are the Jacobians for the two special
logarthmtc integrations and XJA2 is the Jacobian for the standard Gauss
quadrature. The functions ?,, ?2 and ?3 in terms of the new variables n are FLNI
,
FLN2 and FLN3.
(iii) Collocation point at Node C)
This case is similar to the first one with the logarithmic integration variable
being
now
\-z
_
B.100)
Collocation Point
is at 1
Collocation Point
is at 2
Collocation Point
is at 3
Figure 2.16 Geometrical coordinates systems for numerical integration
SUBROUTINE LOCINPQ{XG1,YC1,XG2,YG2,XG3,YG3,GW,NODO)
?
?
PROGRAM 20
?
?
THIS SUBROUTINE COMPUTES THE GW MATRIX MHEK THE COLLOCATION
?
POINT IS ONE OF THE NODES OP THE INTEGRATION ELEMENT.
?
THE COEFFICIENTS ARE COMPUTED BY NUMERICAL INTEGRATION:
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 103
? THE NON SINGULAR PART IS COMPUTED USING STANDARD GAUSS QUADRATURE,
? THE LOGARITHMIC PART IS COMPUTED USING A SPECIAL QUADRATURE FORMULA.
?
COMMON N,L,INP, IPR
DIMENSION GIA0),OMEA0),GILA0),OMELA0),GWC)
?
? DATA FOR THE GAUSS QUADRATURE
?
DATA GI/0.9739065285,-0.9739065285,0.8650633666,-0.8650633666
6,0.6794095682,-0.6794095682,0.4333953941.-0.4333953941,
00.1488743389,-O.1488743389/
DATA ???/0.0666713443,0.0666713443,0.1494513491,0.1494513491
?,0.2190863625,?.2190863625,0.2692667193,?.2692667193,
90.2955242247,0.2955242247/
?
? DATA FOR THE SPECIAL QUADRATURE
?
DATA GIL/0.0090426309,0.0539712662, 0.1353118246,0.2470524162
B,0.3802125396,0.5237923179,0.6657752055,0.7941904160,
GO.8981610912,0.9688479887/
DATA OMEL/0.1209551319,0.1863635425,0.1956608732,0.1735771421
9,0.1356956729,0.0936467585,0.0557877273,0.0271598109,
90.0095151826,0.0016381576/
?
? SET A LOCAL COORDINATES SYSTEM
?
GO TOA,2,3),NODO
1
X3XG3-XG1
Y3=YG3-YG1
X2-OCG2-XG1
Y2=-YG2-YG1
Al=(X3-2*X2)*0.5
B1=X2
A2=(Y3-2*Y2)*0.5
B2=Y2
GO TO 4
2
X3-XG3-XG2
Y3=YG3-YG2
X1=XG1-XG2
Y1=YG1-YG2
A1=X1+X3
B1=X3-X1
A2=Y1+Y3
B2=Y3-Y1
CO TO 4
3
X2=XG2-XG3
Y2=YG2-YG3
X1=XG1-XG3
Y1=YG1-YG3
Al=(Xl-2*X2)*0.5
Bl X2
A2=(Yl-2*Y2)*0.5
B2=-Y2
4
CONTINUE
?
DO 10 J=l,3
10 GW(J)=O.
?
DO 250 I-=l,10
?
? COMPUTE SHAPE FUNCTIONS FOR NUMERICAL INTEGRATIONS
?
F3=0.5*GI(I)*(GI<I)+l.)
F2=1.-GI(I)**2
Fl=0.5*GI(I)*(GI{I)-l.)
FL3=GIL(I) B.*GIL{I)-1.>
FL2=4.*GIL(I>*A.-GIL(I))
FLN3=0.5*GIL(I}
FLN2=1.-GIL(I)**2
FLN1=O.5*GIL(I)*(GIL(I)-1.)
104
Chapter 2 Potential Problems
?
? COMPUTE GEOMETRICAL PROPERTIES
?
GO TOE0,60,70) NODO
?
50 XJA1-SQRT(D*A1*GIL(I>-2*A1+O.5*X3)**2+D*A2*GIL(I)-2*A2+O.5**3)**
? 2)*2
XJA2=SQRT{(A1*GI(I)*2+O.5*X3)**2+(A2*GI(I)*2+O.5*Y3)**2)
XLO=-ALOGB*SQRT((GI(I)*A1+B1)**2+(GI(I)*A2+B2)**2))
S3=FL3*XJA1*OMEL< I)+F3 *XJA2*XLO*OHE (I)
S2-FL2*XJA1*OMEL(I)+F2*XJA2*XLO*OME(I)
S1=FL1*XJA1*OMEL(I)+F1*XJA2*XLO*OKE(I)
GO TO 200
60 XJA1-SQRT(<O.5*B1-A1*GIL(I))**2+@.5*B2-A2*GIL(I))**2)
XJA11=SQRT((O,5*B1+A1*GIL(I))**2+(O.5*B2+A2*GIL(I))**2)
XJA2-SQRT(@.5*Bl+Al*GI(I))**2+@.5*B2+A2*GI(I))**2)
XLO-- O.5*ALOG((GI(I)*Al*0.5+81*0.5)*2+(GI(I)*A2*0.5+B2*0.5)**2)
S3(FIJI1*XJA1+FIJO*XJA11)*OMEL(I)+F3*XJA2*XLO*OME(I)
S2=FLN2*(XJA1+XJA11)*OMEL(I)+F2*XJA2*XLO*OHE(I)
Sl(FIH3*XJAl+FLHl*XJAll)*OMEL(I)+Fl*XJA2*XbO*OME(I)
GO TO 200
70 XJAl-SQRT(B*Al-4*Al*GIL(I)-0.5*Xl>**2+B*A2-4*A2*GIL(IJ-0-5*fl)**
2)*2
XJA2=SQRT(B*A1*GI(I)-O.5*X1)**2+B*A2*GI(I)-O.5*Y1)**2)
XLO=-ALOGB*SQRT((A1*GI(I)+B1)**2+(A2*GI(I)+B2)**2))
S3=FL1*XJA1 *OMEL(I) +F3*XJA2 *XLO*OME(I)
S2-FL2*XJA1*OMEL(I)+F2*XJA2*XLO*OME(I)
S1=FL3*XJA1OHEI,(I)+F1*XJA2*XLO*OHE(I)
?
? COMPUTE GW MATRIX
?
200 GW{3)= GWC)+S3
GWB)- GWB)+S2
GWA) GWA)+SX
?
250 CONTINUE
?
RETURN
END
Routine INTERPQ
This subroutine first reorders the vectors DFI and Fl in such a way that all the
boundary fluxes arc stored in DFI and all the potentials in FI. It then computes
the potentials and fluxes at internal points.
The potential at any interior point is given by
j = O KXj
'= I U ?*????- ? if ***<>>
B.101)
j = 0 KXj
J
j=l lr^ J
Tlie fluxes are given by
{dxl/ J j=| (r \dxi/ )
J
B.102)
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 105
where (he integrals along (he boundary elements arc computed numerically by
calling again the subroutine EXTINPQ.
The listing of INTERPQ is as follows:
SUBROUTINE lKTERPQ(FI,DFI,KODE,CX,Cy,X,y,POT,FLUXl,FLUX2)
?
? PROGRAM 21
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL AND THE FLUXES
? AT INTERNAL POINTS.
?
COMMON N,L,INP,IPR
DIMENSION FIA),DFIA),KODEA),CXA>,CY<1)
DIMENSION XA),VA),POT(I),FLUX1A),FLUX2A)
DIMENSION HWC),CW<3),D4IWC),DQ2VC),DU1NC),DU2VC)
?
? REARRANGE THE FI AND DFI ARRAYS TO STORE ALL THE VALUES OF THE
? POTENTIAL IN FI AND ALL THE VALUES OF THE DERIVATIVE IN DFI
?
NE*N/2
DO ISO I1,NE
DO 170 J*l,3
IF(KODEC*1-3*J)) 110,110,170
110 IF((I-NE).NE.O .OR. J.NE.3) CO TO 125
IF(KODEU)) 114,114.113
113
CH'FI(l)
F1A)*DFIC*I)
DFIC*llCH
CO TO 170
114
DFI<31I'DFII1)
GO TO 170
125 IFII.EQ.l .OR. J.GT.l .OR. KODEI3*1-3).EQ.i) GO TO 130
DFIC*1-2)DFIC*1-3)
GO TO 170
130 CH*FI<2*I-2J)
F1B*I-2J)DFICI-3J)
DF1C*I-3J)'CH
170 CONTINUE
180 CONTINUE
?
? COMPUTE THE VALUES OF ??? POTENTIAL AND THE FLUXES AT
? INTERNAL POINTS
?
IF(L.EQ.O) GO TO SO
DO 240 Kcl.L
POT<K)*0.
FLUX1(K)O.
FLUX2<K)0,
DO 230 I1,NE
CALL EXTINPe(CX(K),CY(K),XB*I-l),YB*I-l),XB*I),YB*l).XB*I*l(
1,YB*1I),HV,GW,DQ1N,DQ2W,DU1U,DU2U,1)
DO 220 J'1,3
IJ22*I-2J
IF(IJ2.GT.B*NE)) IJ2J-2
POT(K)POT(K)OW(J(*DFK3*I-3J)-HW(J)*FIAJ2)
FLUX1(K)FLUXHK)-DU1V(J|*DF1C1-3'J)-DQIW(J)*F1(IJ2)
220 FLUX2(K)-FLUX2(X)*DU2W(J)*DF]C*I-3fJ)-DQ2M(J)*FI(IJ2)
230 CONTINUE
POT(?)POT(K)/<2.*3.1415926536)
FLUXHK) FLUX1 (K)/( 2. 3.14 15926536)
FLUX2<K)FLUX2<K)/B.3.1415926536)
240 CONTINUE
SO RETURN
END
Routine OUTPTPQ
This subroutine prints the results in th following order.
(i), Potentials and fluxes at boundary nodes (fluxes 'before" and 'after" each
node arc printed. Mid-nodes always have the same flux at both sides),
(ii) Internal points potentials.
Chapter 2 Potential Problems
The listing is as follows:
SUBROUTINE OUTPTP(X,Y,FI,DF1.CX.CY,POT,FLUX),FLUX?>
?
? PROGRAM 22
?
? THIS SUBROUTINE PRINTS THE VALUES OF THE POTENTIAL AND ITS NORMAL
? DERIVATIVE AT BOUNDARY NODES. IT ALSO PRINTS THE VALUES OF THE
? POTENTIAL AT INTERNAL POINTS
?
?????? N,L,INP,IPR
DIMENSION X(]|,YAI,F1A).DFIA),CXA),CVU)
DIMENSION POTI).FLUX1A),FLUX2A1
?
NEN/2
WRITE!IPR,100)
100 FORMATI' ',791' 'I//2X,'RESULTS V/2X,'BOUNDARY NODES1//
156X.'POTENTIAL DERIVATIVE'/
29X , *X', 15?,^',12?,'POTENTIAL',6X, 'BEFORE NODE', 6X,' AFTER NODE'/
WRITEJPR,200) X(l),Yil).FI(lt.DFIC*NE),DFl(l)
WRITE!IPR.200| XB).Y<2),FIU>,DFH2),DF1B)
CO 10 1*2,NE
WRITE!IPK,200) XI2*1-1),YB*1-1),FIB*1-1),DFI<3*1-3),DF1C*1-2)
10 WRITE! IPR.200) XB*1).Y<2*I),F1B*I).DFICM-I),DFIC*I-1)
200 FORMAT(&BX,E14.S))
?
IFIL.EQ.O) GO TO 30
WRITE!IPR,300)
300 FORMATt//,2X,"INTERNAL POINTS1,//SX,"X*,15X,'Y',12X,'POTENTIAL',
19X,'FLUX X'.lQJf.'FLUX Y'/)
DO 20 KI,L
20 WRITE!IPR,400)CX(KI,CY<K>.POT!K),FLUX 1<?),FLUX2(K)
400 FORNAT(S|2X,EI4.S))
30 VR1TEUPR.5O0)
SOO FORMAT!' ',t9( '>)
RETURN
ENP
Example 2.4
The problem of an elliptical bar under torsion (figure 2.17(a)) is analysed usin
g
program POQUABI-. Under Saint-Venant type torsion the displacements are
given by
(a)
where 0 is the torsion angle per unit length and ?(?, ?) is the warping function
given by
0
(b)
The boundary conditions are as follows.
Tractions normal to the boundary are identically zero, hence
j|)
(c)
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 10"
,
t ?
?
?
I
s
?
?
?
jo
2
D
?
?
108
Chapter 2 Potential Problems
For the case of an ellipse this becomes,
?? ??-??
(d)
The dimensions were assumed to be a = 10 and b = 5 and values of ? on the
boundary and at two selected internal points (?-, = 2, x2 = 2) and (.v, = 4, .v,
= 3.5)
were computed.
Because of symmetry ? s 0 along the two axes. Thus, only one quarter of the
ellipse needs to be discretized. Ten quadratic elements were used here. Two for
the short semi-axis, four for the long one and four for one quarter of the ellip
se
(see figure 2.17(b)).
The data for this case arc as follows:
ELLIPTICAL SECTION A0) (DATA)
ELLIPTICAL SECTION UNDER TORSION {10 QUADRATIC ELEMENTS)
10 2
0. 0. 1.25 0. 2.5 0. 3.75 0. 5. 0. 6.25 0. 7.5 0. 8.75 0. 10. O.
9.67 1.273 S.814 2.3617 7.7008 3.1898 6.174 3.933
4.7898 4.3891 3.3044 4.719 1.557 4.939
0. 5. 0. 3.375 0. 2.5 0. 1.25
0 0. 0 0. 0 0.
0 0. 0 0. 0 0.
0 0. 0 0. 0 0.
1
1
1
1
0
0 0
0.
-4
-4
-2
0.
0 0
. 0 0.
1 -3.
.8334
.3104
.4411
0 0.
. 0 0.
0 0
379
1 -4
1 -3
1 -1
0 0.
0 0
1 -4
.9447
.4657
.1643
.8334
1
1
1
-4
-2
0.
.3104
.4411
2. 2. 4. 3.5
and the output is given by
ELLIPTICAL SECTION A0) (OUTPUT)
ELLIPTICAL SECTION UNDER TORSION A0 QUADRATIC ELEMENTS)
DATA
NUMBER OF BOUNDARY ELEMENTS' 10
NUMBER OF INTERNAL POINTS- 2
BOUNpARY NODES COORDINATES
NOPE X Y
1
?
3
4
5
6
.OOOOOOOEtOO
. 125?????-?1
.Z500000E+01
.3?50O00EOl
.5000000?-?1
.C250000E401
.????0???-?0
.OOOOOOOEtOO
. OOOOOOOE-tOO
.O0OOO00E4 00
.OOOOOOOE-fOO
.00000O0E00
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 109
7
8
9
10
11
12
13
14
IS
16
17
IB
19
20
.7500000E401
.8750000E401
. 1000000E40Z
.9670000E01
.8814000E401
.7700800E401
.6174000E401
.4789800E40I
.??04400?401
,1SS7OOOEO1
.000O000E400
.OO00000E400
.O00O000E400
.00O000OE400
OOOOOOOE-tOO
.OOOOOOOE400
.OOOOOO0E40O
.1273000E401
.2?61700?401
.3189800E401
.3933OOOE01
4389100E401
.47I9000E401
. 939000?-?1
.5000000E401
.3375OOOEtOl
.2500OOOE+01
.1250000E401
BOUNDARY CONDITIONS
ELEMENT
1
2
?
4
S
6
7
8
9
10
RESULTS
BOUNDARY
FIRST NODEPRESCRIBED
VALUE
.0000000E400
.OOOOOOOE+00
.OOOOOOOEtOO
.00000O0E00
.OOOO0O0E4OO
-,48334OOEO1
-.43104O0E-tOl
-.2441I00E401
.O000O00E4OO
.OOO0OO0E400
NODES
CODE
0
0
0
0
1
1
1
1
0
0
SECOND NODE
PRESCRIBED
VALUE
.OOOOOOOEtOO
,0O000OOE00
,OO000OOE*00
.O0000O0EOO
-.3379OOOEO1
-.49447O0E4O1
-.346S700E401
-.1164300E401
.0000000E400
.O00OO00E4O0
CODE
0
0
0
0
1
1
1
1
0
0
THIRD NODEPRESCRIBED
VALUE
.OOO0OOOE40O
.O0O00OOE400
.0D0000OE4OO
.OOOOOOOE-tOO
-.4833400E401
-.4?1040??4?1
-.2441100E4OI
.OOOOOOOE400
.OOOOOOOE400
.OOOOOOOE400
CODE
0
0
0
0
1
1
1
1
0
0
.OOOOOE400
. 12500E401
.25OOOE+O1
.37S00E4O1
.50000E401
.62S00E401
.TS0OOE40I
.875OOE401
.10000E402
.B67O0E401
.88140E4O1
.770O8E401
.61740E401
.47898E401
.33O44E401
.1SS70E401
.O0O0OE40O
.0OO0OE4O0
.OOO00E4 0O
OO000E400
.OOOOOE+00
.OOOOOE400
OOOOOE400
.OOOOOE400
00000E400
.OOOOOE400
.OOOOOE400
.OOOOOE400
.OOO0OE4O0
.127?0?401
.23817E401
.31898E401
.?93?0?401
.43891E40)
.47I90E401
.49390E401
.S00OOE401
?3750?401
.250OOE401
.12500E401
POTENTIAL
.00O0OE400
.OOOO0E4O0
.00000E400
.00000E400
.O0O00E400
.0OO00E*00
OO000E400
.OO00OE400
.OO000E40O
-.74619E401
-.12506E402
-.14746E402
-.14576E4O2
-.12616E402
-.93634E401
-.46001E401
.00000E400
.OOOOOE400
.O0OOOE400
.0OOO0E4O0
POTENTIAL DERIVATIVE
BEFORE NODE AFTER NODE
.2840SE-03
.7496SE400
.14996E401
.22S02E401
.29999E401
. 37536??1
.44914E401
.S303OE401
.S8638E401
.3379OE+O1
.48334F.401
49447E401
.43I04E401
.34657E401
.24411E401
.11643E401
.00000E400
.2O220E401
.14999E401
.74936E400
-.284DSE-03
.74965E400
.14996E401
.22502E401
.29999E401
.37S36E401
.44914E401
.63030E401
.00000E400
-.33790E401
-.48334E401
-.49447E401
-.43104E401
-.34657E401
-.24411E401
-.?643?401
.?0154?401
.20220E401
.14S99E401
.74936E400
INTERNAL POINTS
POTENTIAL
FLUX X
FLUX Y
.20000E401 .20000E401 -.23990E401 -.12001E401 -.11990E401
.40000E40I .3SO0OE401 -.84019E401 -.21012E401 -.240?1?401
* * * * * *
*** *
*
Results for some representative boundary nodes and the two internal points are
compared with the known exact solution in the following table.
110
Chapter 2 Potential Problems
Boundary node
Potential using 10
quadratic elements
-12.506
-14.576
- 9.363
Exact
potential solution
-12.489
-14.570
- 9.356
x, =8.814)
x2 = 2.361 J
x, =6.174)
r2 = 3.933J
x, = 3.304)
x2 =4.719)
Internal points
*2 = 2.
3
3.5
2.399
8.402
- 2.400
- 8.400
Example 2.5
This example is the solution of the sime elliptical sections as described in fig
ure
2.16(a)) and studied in Example 2.4 but the number of quadratic elements has
been reduced to 5. Il is interesting to see how the simple model gives results w
hich
are in good agreement with the theory.
The input for the five element model (figure 2.17(c)) is as follows:
ELLIPTICAL SECTION E) (DATA)
ELLIPTICAL SECTION UNDER TORSION E QUADRATIC ELEMENTS)
5 2
0. O. 2.5 O. 5.
8.814 2.3617
3.3044 4,719
0. 5. O. 2.5
0 0. 0 0. ? 0.
0 0. ? 0. ? 0.
10. 1 -4.8334
?. 7.5 ?. 10. 0.
6.174 3.933
1 -4.3104 1 -2.4411
? 0. ? ?. ? ?.
2. 2. 4. 3.5
1 -4.3104
1 ?.
2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 111
The corresponding output is given below.
ELLIPTICAL SECTION E) (OUTPUT)
ELLIPTICAL SECTION UNDER TORSION E QUADRATIC ELEMENTS)
DATA
NUMBER OF BOUNDARY ELEMENTS S
NUMBER OF INTERNAL POINTS 2
BOUNDARY NODES COORDINATES
NODE
1
2
3
4
s
6
7
8
9
10
.OOOOOOOEtOO
.2S0OO0OEtOl
.5000000Et01
.750OO0OEt01
.1000000Et02
.8814000Et01
.6174OOOEtOl
.33O44OOE.O1
.OOOOOOOEtOO
.OOOOOOOEtOO
.OOOOOOOEtOO
.OOOOOOOEtOO
.00000001*00
.OOOOOOOEtOO
.OOOOOOOEtOO
.23617???-?1
.3933O0OEt01
.4719O0OEt01
.SOOOOOOEtOl
2S0OOOOEtOl
BOUNDARY CONDITIONS
ELEMENT
1
2
?
*
?
FIRST NODEPRESCRIBED
VALUE
.00OO000EO0
.0OO00O0E0O
.OOOOOOOEtOO
-.43104001401
.0OO000OE0O
CODE
0
0
1
?
SECOND NODE
PRESCRIBED
VALUE
.OOOOOOOE+00
.OOOOOOOEtOO
-.4833400E401
-.2441IOOEO1
.OOOOOOOE+00
CODE
0
0
1
1
?
THIRD NODEPRESCRIBED
VALUE
.O000OO0EOO
.OOOOOOOE-.00
-.4310400E401
.OOOOOOOE'OO
.OOOOOOOEtOO
CODE
0
0
1
1
0
RESULTS
BOUNDARY NODES
.000O0EOO
.25OOOE.01
.50000E01
.750O0EOl
.10000E+02
-88140E+01
.61740E*01
.33O44EOS
.OOOOOEfOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.23617E*01
.39330E*01
.4 719OE-.O1
.50000E+0)
.ZSOOOEtOl
POTENTIAL
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
.OOOOOEtOO
-.12779?tO2
-.14839E+02
-.9435OEtOl
.OOOOOEtOO
.OOOOOEtOO
POTENTIAL DERIVATIVE
BEFORE NODE AFTER NODE
.103B8E-01
.15215EtOl
.297S7EtOS
,46612E0i
.5162SEt01
.43IO4?tOl
.24411EtO)
.OOOOOEtOO
.15277EtOl
-.1038SE-01
.!S21EtOl
.297S7EtOl
.46612EtOl
.OOOOOEtOO
,48334EtOl
-,431O4EtOl
-.244UEtOi
.3OOO4EtOl
15277EtOl
INTERNAL POINTS
X
,20000Et01
.OOOOEtOl
.20000E401
,35000Et01
POTENTIAL
-.2430SEt01
-.S4718EtOl
FLUX X
-.1213OE+O1
-.2U!lEt01
FLUX Y
-.12163E*O1
-.23S9OEtOl
* * *
* * *** ** *
* * *** **
112
Chapter 2 Potential Problems
The following table compares the 5 element solution with the exact solution.
Boundary node
x, =8.8141
x2 = 2.361 J
x, =6.1741
x2 = 3.933 J
x, =3.3041
x2 = 4.7I9j
. Internal points
x,=4 1
Potential using 5
quadratic elements
-12.779
-14.839
- 9.435
- 2.431
- 8.472
Exact
potential solution
-12.489
-14.570
- 9.356
- 2.400
- 8.400
As an exercise, the reader can run the same problem using programs
POCONBE and POLINBE for constant and linear elements and compare results
with those shown in the above table and in Example 2.4.
2.10 Computer Code for Multiboundary Problems (POMCOBE)
In engineering practice many problems have more than one surface as shown in
figure 2.18, with internal and external boundaries. Both types of boundary can b
e
differentiated by identifying the direction of the normals. This can easily be d
one
in two dimensional problems by adopting the rule that the numbering on the
external surface is done counterclockwise and the one on the internal surface is
carried out in the clockwise direction. From these rules the normal will be well
defined in the computer code.
The following computer code is based on the constant element code (POCONBE)
of section 4, but while the previous code was applicable to problems with only
one surface the following listing applies for muitisurface cases.
Main Program
Similar to program I already described in POCONBE, but now the Common
statement is replaced by
COMMON N,L,NCE),M,LECJMP
2.10. Computer Code for Multiboundary Problems (POMCOBE)
113
(a) Turbine blade with cooling holes
(b) Set ol tunnels
I
(c) Potential flow around several obstacles
Figure 2.18 Problems with more than one surface
where M defines the number of different surfaces and NC stores the last node of
each different surface. The dimensions of NC allow for 5 different surfaces in t
he
present listing but this can easily be extended by the user if required.
114
Chapter 2 Potential Problems
The listing is us follows:
?
?
?
PROGRAM ???????
?
? PROGRAM 23
?
?
? THIS PROGRAM SOLVES TWO DIMENSIONAL (PO)TEMTIAL PROBLEMS
? WITH (M)ULTIBOUNDASY DOHA INS USINC (CONSTANT (B)OUNDARY (E)LEMENTS
?
?
CHARACTER* 10 FZLEIN.FILEOUT
?
PI KENS I OS XI10) ) ,Y< 101 ) ,XM( 100) ,YM( 100) ,FI< 100) .DFIOOO)
DIMENSION KODEUOO),CXB0),CYB0),POTB0),F1.UX]B0).FLUX2B0)
?
COMMON/MATG/ 6A00,100)
COMMON/MATH/ NA00,100)
COMMON N,L,NC(S),M,INP,IPR
?
? SET MAXIMUN DIMENSION OF THE SYSTEM OF EQUATIONS (NX)
? (THIS NUMBER MUST BE EQUAL OR SMALLER THAN THE DIMENSION OF XH, ETC..)
?
NX*100
?
? ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES
?
INP'S
IPR=6
?
? READ NAMES AND OPEN FILES FOR INPUT AND OUTPUT
?
WRITE!>,' (A) ') ' NAME OF INPUT FILE (NAX. 10 CHART.)'
READ<,' (A) 'IFILEIN
OPENU?,FILE*F1LEIN,STATUS*'OLD')
NAME OF OUTPUT FILE (NAX. 10 CHART.)'
WRITE(>,# (A) ')
READ)*,' (A) ')FILEOUT
OPEN!IPR,FILE>F1LEOUT,STATUS*"NEW")
?
? READ DATA
?
CALL INPUMPC<C.X,CY,X,Y,KODE,FI)
?
? COMPUTE H AND G MATRICES AND FORM SYSTEM (A X F)
?
CALL CHMAMPCIX.Y.Xtf.VM.C.H.Fl.DFl.KODE.NX)
?
? SOLVE SYSTEM OF EQUATIONS
?
CALL SLNPD(G,DFI,D,N,NX)
?
? COMPUTE POTENTIAL VALUES AT INTERNAL POINTS
?
CALL INTEMPC(FI.DFI,KODE,CX,CY,X,Y,POT,FLUX1,FLUX2)
?
? PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS
?
CALL OUTPMPC(XH,YM,FI,DFI,CX,CY,POT,FLUX1,FI-UX2)
?
? CLOSE INPUT AND OUTPUT FILES
?
CLOSE (INP)
CLOSE (IPR)
STOP
END
2.10. Computer Code for Multiboundary Problems (POMCOBE)
115
Routine INPUMPC
The input required is the same as in Program 2 of POCONBE with the exception
of M and NC which are read in the same line as N and L (i.e. number of elements
and number of internal points where the function u is required).
The listing is as follows:
SUBROUTINE INPUMPC(CX,CY,X,Y,KODE,FI)
?
? PROGRAM 24
?
CHARACTER*80 TITLE
COMMON N,L,NCE>,M,INP,IPR
DIMENSION CXA),CYA),XA),YA),KODEA),FIA)
?
? N= NUMBER OF BOUNDARY NODES (=NUMBER OF ELEMENTS)
? L= NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED
?
WRITE(IPR,1OO)
100 FORMAT(' ',79('*'))
?
? READ JOB TITLE
?
READ(INP,'(A)') TITLE
WRIT?(IPR,'(A)') TITLE
?
? READ NUMBER OF BOUNDARY ELEMENTS,NUMBER OF INTERNAL POINTS,
? NUMBER OF DIFFERENT BOUNDARIES AND LAST NODE OF EACH BOUNDARY
?
READ(INP,*)N,L,M,(NC(K),K=1,M)
HRITE(IPR,300)N,L
300 FORMAT(//' DATA'//2X,'NUMBER OF BOUNDARY ELEMENTS =',I3/2X,'NUMBER
1 OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED =',I3)
?
IF(ML0,40,30
30 WRITE(IPR,999)M,<NC(K),K=1,M)
999 FORMAT</2X, 'NUMBER OF DIFFERENT BOUNDARIES ',I3/2X,
l'LAST NODE OF EACH BOUNDARY =',5A3,','))
?
?
?' READ COORDINATES OF EXTREME POINTS OF THE BOUNDARY ELEMENTS
? IN AKKAVS X AN Y
?
40 WRITE(IPR,500)
500 FORMAT(//2X, 'COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELE
1MENTS',//IX,'POINT',7X,'X',15X,'Y')
READ(INP,*) (X(I),Y{I),I1,N)
DO 10 1=1,N
10 WRITE(IPR,700I,X(I),Y(I)
700 FORMATBX,I3,2BX,E14.5))
?
? READ BOUNDARY CONDITIONS IN FI(I) VECTOR, IF KODE(I)=0 THE FI(I)
? VALUE IS A KNOWN POTENTIAL;IF KODE(I)-1 THE FI{I) VALUE IS A
? KNOWN POTENTIAL DERIVATIVE (FLUX).
?
WRITE(IPR,BOO)
800 FORMAT(//2X, 'BOUNDARY CONDITIONS'//2X, 'NODE' , 6X, 'CODE' , 7X, 'PRESCRI
1BED VALUE')
DO 20 1=1,N
READ(INP,*) KODE{I),FI(I)
20 WRITEdPR, 950I, KODE(I) , FI (I)
950 FORMATBX,I3,9X,I1,8X,E14.5)
?
? READ COORDINATES OF THE INTERNAL POINTS
?
IF(L.EQ.O) GO TO 50
READ(INP,*)
50 RETURN
END
116
Chapter 2 Potential Problems
Routine GHMAMPC
In this routine the COMMON needs to be changed as in the MAIN program. In
addition some extra commands have been included to differentiate the points on
each of the surfaces. These arc required in order to compute the mid-point
coordinates XM and YM. Each surface has to close and the last node of each
surface is in the mid-point between the last extreme point to the First point on
that surface.
The listing of GHMAMPC is now as follows.
SUBROUTINE GHMAMPC(X,r.XM.VH,G,H,FI,DFl,XODE,NX)
?
? PROGRAM 29
?
? THIS SUBROUTINE COMPUTES THE 0 AND N MATRICES
F
? AND FORKS THE SYSTEM OF EQUATIONS A X
?
COMMON N.L.NCfS).H.INP.IPR
DIMENSION XA),VA) XHt1),YM<I),FI(I),KODEA)
DIMENSION DFIA),G(NX,NX),H(NX,NX)
?
? COMPUTE THE NODAL COORDINATES AND STORE IN ARRAYS XM AND YM
?
DO 10 I1,N
10
IF(M-OI5,15,12
12
XM(NC(l))>(X(NC(l))X(l))/2
TM(NCA)>(Y(NCA))*YA))/*
DO 13 K2,H
XM<NC(K))<X(NC<K)>*X(NCOC-l) +
13
YH<NC(K))*(Y(NC(X)>Y(NC<K-lLl>)/2
?
? COMPUTE THE COEFFICIENTS OF ? AND H MATRICES
?
15
DO 30 II,N
DO 30 J'l.N
IF(M-1I6,16,17
17
IF(J-NC<1)I9,18,19
18
KX1
GO TO 23
19
DO 22 X*2.M
IF(J-NC(K)J2,21,22
21
KKNC(K-l)i
GO TO 23
22
CONTINUE
16
JCKJ1
23
IF(l-JJ0,25,20
20
CALL EXTINPC(XM(I),TM[l|,X(J),Y(-5).X(KK),-y(KK).H(I,J) .C(I.J)
l,DQl,Dq2,DUl,DU2,O)
SO TO 30
25 CALL LOCINPCIX(J),Y(J),X(KX),Y(KK),GA,J)>
H(I,J)>3.1415926
30 CONTINUE
?
? REORDER THE COLUMNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE
? WITH THE BOUNDARY CONDITIONS AND FORM SYSTEM MATRIX A WHICH
? IS STORED IN G
?
DO 55 Jl,N
IF(KODE(J)M5,5S,40
40 DO 50 I-1,N
CHGA,J)
G(I,J).-H(I,J)
K(I,J)-CH
50 CONTINUE
55 CONTINUE
?
? FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN DFI
?
2.10. Computer Code for Mulliboundary Problems (POMCOBE)
117
DO 60 1*1,N
DFIA)*O.
DO 60 J*1,N
DFI(I|>DFI(ILH(I,J)*FI(J)
CO CONTINUE
RETURN
END
Routine EXTINPC
As in POCONBE (Program 4).
Routine LOCINPC
As in POCONBE (Program 5).
Routine SLNPD
As in POCONBE (Program 6).
Routine INTEMPC
This routine varies from Program 7 by a few statements to take into account the
different surfaces.
SUBROUTINE INTENPC<FJ,DFI,RODE,CX,CY,X,Y.POT,FLUX1.FLUX2)
?
? PROGRAM 26
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL
? AND THE POTENTIAL DERIVATIVES (FLUXES) AT INTERNAL POINTS
?
COMMON N,L,NC<S),H,INP,IPR
DIMENSION FIU >.DFIA),KODEA),CXA),CYA),XA),YA)
DIMENSION POTA),FLUX1A).FLUX2A|
?
? REARRANGE THE FI AND DFI ARRAYS TO STORE ALL THE VALUES OP THE
? POTENTIAL IN FI AND ALL THE VALUES OF THE DERIVATIVE IN DFI
?
DO 20 Il,N
IFIKODE(I)) 20,20,10
10 CH'FIll)
DFHIJCH
20 CONTINUE
?
? COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
IFIL.E9.0) GO TO 50
DO 40 Kti.L
POT(K)0.
FLUXHKHO.
FLUX2IKM0.
DO 30 J],N
IF<M-1J8,28.22
22
IF(J-NC(I)J4,23,24
23
KK<I
GO TO 29
118
Chapter 2 Potential Problems
24
DO 26 LK=2,M
IF(J-NCILKI26,25,26
25
KK=NC(LK-1)1
CO TO 29
26
CONTINUE
28
KKJH
29
CALL EXT1NPC(CX(K),CY(X),X(JI.V(J),X(KK),Y(KK),A.B
1.D41.D92.DU1,DU2,1)
POT(K)POT(K)DFIU)*B-F1(J)*A
FLUX1(K)FLUX1<K)DF1< J)DU1-FI( J)MJ1
30
FLUX2(X)>FLUX2{K>*DFI<J)*DU2-FI(J)*DQ2
POT(K)fP0T(K>/B.3.1415926)
FLUX MX). FLUXHX>/<2.*3.1415926)
40 FLUX2<K)> FbUX2{K)/<2.3.1415926)
SO RETURN
END
Routine OUTPMPC
Siimc as Program 8 but with Ihe new COMMON.
* .
>.-.. .
~
C*-~""~~~~*"
SUBROUTINE OVTI>MPC(XN,YM,FI ,DF1,CX,CY,POT.FLUX1,KLUX2)
?
? PROGRAM 27
?
? THIS SUBROUTIHE PAINTS THE VALVES OF THE POTENTIAL AND ITS NORMAL
? DERIVATIVE AT BOUNDARY NODES. IT ALSO PRINTS THE VALUES OF THE
? POTENTIAL AND THE FLUXES AT INTERNAL POINTS
?
COHNON N,L,NCF),H,INP,IPR
DIMENSION XM< 1) , YHU ) ,FI ( > ) .DF1 A), CX( 1 ) ,CY( 1)
DIMENSION POTI1),FLUX1A),FLUX2I1)
?
VRITE(IPR.IOO)
100 FORMAT) ' ',79( *1)//1X,'RESULTSV/2X,'BOUNDARY NODES V/8X, 'X' , 15
lX.'Y',1 ??,'POTENTIAL1,??,'POTENTIAL DERIVATIVE'/)
DO 10 I'l.N
10 WRITE! IPR.200) XM(I),rM(I),FIU).DFI(I)
200 FORMATDBX,E14.5)>
?
IF(L.EQ.O) 00 TO 30
WRITEUPR.300)
300 FORMAT!//,2X,'INTERNAL POINTS*,//8X,"X*,I5X,'V,1??,'POTEKTIAf,
19X,'FLUX X'.IOX.'FLUX Y'/>
DO 20 K>1,L
20 WRITE!1PR,400)CX(K),C(K),POT(K),FLUX 1(K),FLUX2(K)
400 FORMAT(Si2X,E14.5))
30 WRITE)IPR,500)
500 FORMAT!' ',79('*'))
RETURN
END
2.11 Boundary Elcmeats for Three Dimensional Problems
The elements used in three dimensional problems are surface elements which cover
the boundary of the body (figure 2.19). They are usually of two types; triangula
r
or quadrilateral and both can be flat or curved. The functions ? and q and those
used to describe the geometry can be constant over the element, vary linearly,
being second order functions and others which produce a curved element. While
the development of constant or Hat elements is comparatively simple, curved
elements are more important in three dimensions as they can follow better the
geometry of actual engineering components and hence they will be described in
detail here.
2.11. Boundary Elements for Three Dimensional Problems
119
Figure 2.19 Some three dimensional applications (lower figure by courtesy of
British Aerospace pic)
To study curved elements first we need to define the way in which we can pass
from the xlx2xi global cartesian system to the ?,?2, ? system defined over theelement, where ?,?2 are oblique coordinates and r\ is in the direction of the no
rmal
(see figure 2.20).
120
Chapter 2 Potential Problems
Figure 2.20 a) Triangular curved elements for three dimensional problems
The transformation for a given function - say u - is related through the
following,
du
du
??,
Wi
??.
??
???
__
??2
??
??2
??3
???
??3
??
?? '
?^
??
??2
??
B.103)
2.11. Boundary Elements for Three Dimensional Problems
\ -*
121
3
*2
Figure 2.20 b) Quadrilateral curved elements for three dimensional problems
where the square matrix is the Jacobian or J. Hence
?? ?
??
??
??
?;
?? '
dxt
??
??2
??
B.104
122
Chapter 2 Potential Problems
The inverse relationship is then given by
du
dxx
du
cx2
Du
i f
1 ,
, J ,
du '
du
??
du
B.105)
Transformations of this type allow us to describe differentials of volume or
surface in the cartesian system in terms of the curvilinear coordinates. For ins
tance
a differential of volume Q can be written as,
<IQ = Magnitude)-- x ~f ~~ / <tf, d?2
dij
A differential of area instead will be given by
B.106)
B.107)
where G is a reduced Jacobian and |G| is simply the magnitude of the normal
vector fa i.e.
= x
dq
B.108)
where
Notice that the values of gt, g2 and g3 are given by
c2 dx3 dx2 dx3^
(??3 ??% ??? ??
dxj fa, _ Sx2 ??
Hence the magnitude of |G| is given by
B.109)
B.110)
2.11. Boundary Elements for Three Dimensional Problems 123
These relationships can be used to integrate any of the boundary integrals terms
such as,
JiiV/r or J4*<flT
B.111)
?
?
which now become
J u*q\G\ d^ dt2; 1 uq*\G\ ilv dl2
B.112)
A whole range of quadrilateral and triangular elements can be defined. For
quadrilateral cases one prefers to use Lagrangian type of elements (i.e. the one
s
which have in some cases nodes inside) as such elements give better numerical
results when used with point collocation. This is simply because the collocation
points arc belter distributed in these cases. Table 2.1 shows some of the elemen
ts
which can be used.
2.12 Poisson's Equation
Many practical applications are governed by the Poisson rather than Laplace
atio
a?
.?
|
?2
.2
. ~
5 -?C
?
.?
?
J
rder
?
2
3
2
af/o.
??/
f?r
?;
?
.2
net
.,_ ?
II
?
1
I
}|
n
*. ,
"-!
II
II
1
+
II
?4
f4l
|
1
"?*
-^
V
i
"
i
MM
II
<u
3
+
1
II
J M
l+
+
II
? ^^^^^
J
3
Oi
<o
7
tij. i
l
aa
cond
der
??
-2
5
3
no
pol
ter
,C
??
" u.
|
1
Cl
?
-? _
-~ >IJ,
1 "*
II
II '^J1
1 'I
" ^1"
ii II
T -?
i ~^*
??" ?"
II II
^ ^
1
n
"J(I ~
ir "
~ i
"" 1
1 1
i i " f
' tl It
--?1
?
3
*4
oo
1
If
2.12. Poisson's Equation
125
Aflcr substituting the Laplace fundamental solution ?* and grouping all boundary
terms together (i.e. in ? = ?, + ?2), one obtains
< V + f uii* (ir + $hu*du = $(iu*dr
B.116)
r
r
Notice that although the b functions are known and consequently the integrals
in Q do not introduce any new unknowns, the problem has changed in character
us we need now to carry out integrals in the domain as well as on the boundary.
Regions of integration called cells can now be employed to compute the domain
integral in B.116) (figure 2.21). One usually applies a numerical integration
scheme such as Gauss. In this case for each position of the singularity at a bou
nboundary point i, the integral in B.116) can be written as,
= t (t
B.117)
where ? arc the different cells (e varies from I to M, where M is the total numb
er
of cells describing i\ domain), \\\ arc the integration weights, the function (h
u*)
needs to be evaluated at r integration points on each cell, Qe is the area of th
e
cell V\ ?>' is the result, different for each '? position of the fundamental sol
ution,
where / is one of the boundary nodes.
Hence equation B.116) now becomes
H 8)
?V + ? iVhi> + D'=t ? V B
or in matrix form
HU + D = GQ
B.119)
Notice that the domain integrals need to be computed as well when calculating
any values of potentials or fluxes at internal points. Hence,
'= ? C'V- I H'V-D1
B.120)
i= i
J=i
where /' is now an internal point, at which the singularity is applied.
Concentrated sources are very simple to handle in boundary elements. They
are a special case for which the function b at the internal point T becomes,
b = QlA'
B.121)
where Q' is the magnitude of the source and A1 is a Dirac delta function whose
integral is equal to 1 at the point / and zero elsewhere. Assuming that a number
12O
Chapter 2 Potential Problems
of these functions exist one can write.
p
<V + J uq* <1? + f bu* du + Y. (Q'm*1) = J qu* <1?
B.122)
?*' is the value of the fundamental solution at the point /. P are the number
of concentrated forces within the domain.
Another way of dealing with volume sources is to transform the domain integral
into equivalent boundary integrals. This is possible when the function h is harm
onic
(although the approach can in principle be extended to harmonic, or bi-harmonic,
etc.). Harmonic means that b obeys the following equation,
V2ft = 0
B.123)
To effect the transformation - which is basically an integration by parts - we n
eed
to express the fundamental solution ?* in function of some derivatives. This is
done by proposing a function v* such that,
h* = W B.124)
(Notice that the cells
are rather arbitrarily
defined)
cell
Figure 2.21 Boundary elements and internal cells
2.12. Foisson's liquation
and then writing Green's identity, in the form
J (WV - v*V2h)<ICl=Ub~- v* )</?
B.125)
a
r \ On dn)
which if V2fr = 0 reduces to,
r/>H*</Q = j(/)~-i>*^W B.126)
u
i \ dn dn)
Hence we have effectively reduced the domain integrals to two different boundary
integrals.
The function v* required in this case is simply the fundamental solution of the
biharmonic equation as,
V2M* = V2(W) = W = -?' B.127)
which for two dimensions is a well known fundamental solution used in plate
bending, i.e.
B-?)
and for three dimensions
p* = -ir
B.129)
8?
Notice that for two dimensions v* satisfies Laplace's equation as follows,
f^
B.130)
? vr \ or) 2n r
and for three dimensions one finds that,
Example 2.6 Results for the Poisson Equation Using Internal Cells
The equation V2 = -2 was solved for the geometry shown in figure 2.22 with
the homogeneous boundary condition ? = 0 and using linear elements.
128
Chapter 2 Potential Problems
Figure 2.22 Elliptical section discretized with Boundary Elements and internal
cells
Notice that the domain has been discretized with a total of 48 internal cells.
Results presented in table A below,demonstrate that the solution is accurate.
Table A Cell results for Poisson Problem
Node
17
18
19
20
29
30
31
X
1.5
1.2
0.6
0.0
0.9
0.3
0.0
Y
0.0
-0.35
-0.45
-0.45
0.0
0.0
0.0
Cell
0.331
0.401
0.557
0.629
0.626
0.772
0.791
Exact
0.350
0.414
0.566
0.638
0.638
0.782
0.800
The reader can verify these results by using program POLINBE, after having
added an appropriate subroutine to compute the vectors D as described in the
previous section.
2.13 Orthotropy and Anisotropy
Up to now we have considered problems with isotropic materials, i.e. those for
which the properties arc the same in all directions. We will study those materia
ls
for which the properties vary in different directions. If one finds the directio
ns of
orthotropy (Figure 2.23) the governing equation for a two dimensional problem
2.13. Orthotropy and Anisotropy
129
Figure 2.23 Orthotropic medium
can be written.
2
dyj
?2
?1' V2 directions ol
orthotropy
B.132)
A-,- is the medium properly coefficient in the direction of orthotropy i. For th
ree
dimensional cases the equation is
, ?2?
B.133)
The fundamental solution corresponding to the above equations can be found
by transforming the y,-system of coordinates into a Zj system in which the gover
ning
equations B.132) or B.133) become a Laplace equation without the kt coefficoefficients. This is achieved by using the following transformation,
!
B.134)
The fundamental solution of the Laplace equation in the z-t system is known
and can then be back transformed to the Xj coordinates. This gives for the two
dimensional case the following solution:
?* =
B.135)
130
Chapter 2 Potential Problems
where r is now
fi
{2.136)
where y,, y2 are tne coordinates of the point under consideration and y\y'2 are
those of the node on which the fundamental solution is applied.
The corresponding normal flux is
q* = kt -----n,, + k2 - nv2
B.137)
<?y,
r7y2
where nyl and n,,2 are the direction cosines of the normal ? to the surface with
respect to y, and yz respectively. Similarly for the actual flux we have
</ = *,-- +*,-- ny2
B.138)
For a fully anisotropic media, governing equation B.132) becomes,
, 02u ., rl2u ,
, +2fc12 _ + fc2 =0
B.139)
avf
axtdx2 dxj
where the fc4 coefficients represent the terms of the properties tensor. The
fundamental solution can now be expressed
1 In- B.140)
where \ko\ is the determinant of the medium conductivity and s is the inverse of
the ? matrix, ie the resistivity matrix
"^12 KH J
B-141)
The r distance in equation B.140) is now
? = 1 *,, [jc| - x,]2 + 2st2 [x\ - x,][x2 - x J 4- sn[x2 - x2]2) ? B.142)
The normal flux is
/ ??* ??*\ ( ??*
??\
4*- ?-,,, -+* -" ? + *,2~- + *2?-?2 B-143)
V <>x, ??2/ \ dxt
dxj
and similarly for q.
The fundamental solution for the three dimensional orthotropic case is
B.144)
2.13. Orthotropy and Anisotropy
where r is now
131
? = <? ?/. -??2 + ? ?? ~??? + ~?-??? \ B-145)
Similar considerations as for two dimensional cases apply for the fluxes q* and
q.
2.14 Subregions
In certain cases the region under study may be piecewise homogeneous and then
the boundary element procedure can be applied to each subregion in turn as if
they were independent of each other. The final set of equations for the whole re
gion
can then be obtained by assembling the set of equations for each subregion using
compatibility of potentials and fluxes between the common interfaces.
Consider for instance the two subregions shown in figure 2.24 one called ?1
and the other Q2. Over subregion fi1 we define,
U\ 01; Nodal potential and fluxes at the external boundary ?1
U|\Q|'; Nodal potential and fluxes at the interface ?, considering it
belongs to Q|
Figure 2.24 Piecewise homogeneous body consisting of two subregions
132
Over the other subregion, Qz, one can define,
Chapter 2 Potential Problems
U2,Q2; Nodal potential and fluxes at the external boundary ?2
?2< Qn Nodal potential and fluxes at the interface ?, considering it
belongs to ?l2.
The system of equations corresponding to ?, can be written as
B.146)
and the one for subregion Q2 gives
B.147)
The compatibility and equilibrium conditions on the interface ?, can be expresse
d as
U,' = U2
B.148)
B.149)
and
If one calls the potentials at the interface U, and adapts on the same interface
the
fluxes of Qj as reference values (which is equivalent to say that the normal on
the
interface is the normal to fi.) one has,
and
U, = U,1 = U,2
B.150)
B.151)
These conditions can be introduced in equations B.146) and B.147) which
can now be written together as follows,
U1
?1 H,1 Oil I ??1 G,1 0]
o H,2 H2JI '( j_0 -G,2 G2J
V2
Q1
Q.
Q2
B.152)
As !J| and Q, are unknown at the interface the above system is frequently writte
n as
u'1
??1 H/ -G,1 0 1
LO H,2 G,2 H2J
Q.
u2
o
B.153)
2.14 Subrcgions 133
Notice that we still have to apply the relevant boundary conditions on the exter
nal
surface of the region, i.e. ?1 and ?2. It should also be noticed that the matrix
on
the left hand side of equation B.153) is square and that the one on the right ha
nd
side is not.
It is interesting to point out that when more interfaces are included the system
of equations may tend to be banded or have a large number of zero submatrices,
which can result in an A matrix which is computationally more efficient. For thi
s
reason most boundary clement codes offer nowadays subrcgions, particularly for
analysing three dimensional problems.
2.15 Hclmholtz Equation
Another useful equation in potential problems is the so called Hclmholtz or wave
equation. Its time dependent version is
V2u + ?2? = 0 in ?2
B.154)
where V2 is the two or three dimensional Laplacian and )} is a positive and know
n
parameter. Let us consider that the boundary conditions arc the normal
(i) Essential conditions ? = ? on ?,
B.155)
q on ?2
(ii) Natural conditions q
Mixed boundary conditions can easly be incorporated in the formulation and arc
important in many practical problems. We will however restrict the discussion
now to the above two conditions for simplicity.
Th,e corresponding weighted residual statement for B.154) and B.155) is
f (V2u + /2m)m* du = J {q - q)u* </? - J (u - u)q* </? B.156)
it
??
?|
Integrating twice by parts one obtains.
J (W + A V )u du = - J qu* </? - f qu* dT
+ | uq* </? + | uq* dV B.157)
or in more compact form
nvV + AVMn=-J</M*</r + jM</*</r B.158)
ii
?
?
134
Chapter 2 Potential Problems
where ? = ?, + ?2.
The fundamental solution ?* for the Helmholtz equation now needs to satisfy
V2w* + >.V + ?( = ?
B.159)
For two dimensions this results in
?*=' Wi,1'??) B.160)
4
and
/) ;
q* = = - : H\l)tir) B.161)
dr 4
where r is as usual the distance from the source point to the point under consid
erconsideration, i = yf^-1, H|>
is the Hankel function of the first kind and zero
order and
H\u a Hankel function of the first kind and first order, i.e.
?) = J0()jt) + /T0(/r) B.162)
B.163)
where J and V are Bessel functions of first and second kind, with the subscripts
indicating their order.
The three dimensional fundamental solution is
ar
B.164)
4??
and
?\ ? /
JL ( . + i}\ e"r
B.165)
4??\
/
2.16 Axisymmetric Formuiation
A way of dealing with the formulation for axisymmetric problems is by integratin
g
the three dimensional solution.
Assume first that all boundary values have axial symmetry and consequently
all domain values are axisymmetric. We can write the governing integral equation
B.42)in terms of the cylindrical polar coordinates (p,0,Z) given in figure 2.25a
s,
c'V + Jm J q*depdt=\q I u*dOpdt B.166)
? ?
? ?
2.16 Axisymmetric Formulation
135
where ? is the generating body created by the intersection of the three dimensio
nal
? surface with the part of the plane defined by p and Z positive.
The three dimensional fundamental solution can be written in cylindrical
coordinates
i
Am 4??{?, 0, Z)
B.167)
and then integrated with respect to 0, which gives the type of ring source solut
ion
required by axisymmetric problems, i.e.
'(
where: m = -"
b
(a + bI'2
B.168)
B.169)
Figure 2.25 Geometrical definition, generating area and boundary contour of
solid of revolution
136
Chapter 2 Potential Problems
K(m) = is the complete elliptical integral of the first kind.
The subscript T refers to the position of the fundamental solution. Notice that
unlike the 2 or 3 dimensional cases, the axisymmetric fundamental solution can
not be written simply as a function of the distance between two points but also
depends on the distance from the source T, the point under consideration to the
axis of revolution.
A different formulation of the axisymmetric problem will be presented in section
3.9 where the elliptical integral is avoided by direct numerical integration of
the
three-dimensional fundamental solution.
2.17 Indirect Formulation
Up to now we have been using the so-called direct formulation which is nowadays
the one commonly associated with boundary elements. In the past it was frequent
to associate boundary integral solutions with what is now called the 'indirect'
approach. In this formulation the boundary solutions arc obtained by using sourc
es
or sometimes dipolcs and it is interesting to point out that this can be interpr
eted
as a particular case of the more general direct approach.
Considering for instance the Poisson equation, i.e.
V2u + b = 0 infi
B.170)
The weighted residual formulation applied on B.170) produces the direct method
integral statement, i.e. for any point in the body.
vf + J ?* dV = J u*q <1? - J bu* ilu
B.171)
?
r
ii
If we now call the region external to ? by ft' and assume thatu' is (he solution
to the Laplace equation, V2' = 0 within ?', one can write the following statement
,
f q*u' A?-\ u*q' dV = 0 B.172)
r
?
where the point V has been assumed as external to Q' (internal to Q) and the
tractions q and q* have been referred to the normal n to the internal domain so
that the two values of q* in B.171)and B.172) become identical. One can now
specify u' as the solution in Q' which generates potentials around ? identical t
o
those of our initial problem, in Q, i.e.
???' on ?
B.173)
We can now add equation B.153) and B.154) to give
_ $u*{q-q')dr + lbu*dCl = Q B.174)
?
n
2.17 Indirect Formulation
137
or,
ul=fn*<rdr-fbumiin
B.175)
?
n
a = q
q' represents the initially unknown density distribution of u* over ?
paragraph above and called the Multiple Reciprocity Method [ 12] and the other
on the idea of using a series of particular solutions and called Dual Reciprocit
y
Method [13] in the literature. They will be described in the next sections, afte
r
a brief description of the use of particular solutions as an introduction to Dua
l
Reciprocity.
(i) The Use of Particular Solutions
An obvious way of solving equation B.180) without domain integrals is by changchanging the variables in such a manner that these integrals disappear. This can
be
attempted by adding a particular solution to a new variable.
To illustrate the procedure consider the Poisson equation B.179) with boundary
conditions such as
Essential conditions u = ? on ?,") _
I i
1 | T lj (Z. Io3)
Natural conditions q = q on T2)
Assume now that the potential ? can be written as,
? = ? + ?
B.184)
2.18 Other Approaches for the Treatment of Domain Integrals
139
where ? is a particular solution of the Poisson equation, such that
V2& = h B.185)
One can now write the domain term in equation B.180) as follows,
J bu* <IQ = J (V2i))* du
B.186)
12
U
Integrating by parts this expression one finds the following relationships,
} bu* du = J (V2u)u* dQ
= f fi(W)dn + J ?* q dT - J q*u dT
B.187)
u
?
?
?
- ?'
where ? = .
Taking into consideration the special properties of the fundamental solution (se
e
section 2.2) one can write the right hand side term in B.187) as follows,
f bu* dQ = -<V + J u*q dV - J q*u dV
B.188)
?
r
Substituting B.188) into B.180) one finds the following expression,
(''" if ,i*u </? - J </?* </?
i
B.189)
= c'ft1" + f <i*t, dT - J i)u*
?
?
Notice that now all integrals need to be computed only on the ? boundary. EquaEquation B.189) can also be written in a more compact form as function of the ne
w
variable ?, i.e.
i4t + J q*u dT = J f/u* dV
B.190)
where ? ?
? defines the new variable.
The main difficulty with particular solutions is that they are difficult to find
in
many cases and cannot usually be applied to time dependent or non-linear
problems.
140
Example 2.7
Chapter 2 Potential Problems
This example describes how to apply particular solutions to the elliptical secti
on
described in example 2.6. (Figure 2.22).
Here the Poisson equation V2u == 2 is solved using the solution procedure
-0.045
0.000
? = ? + ?
0.348
0.419
0.574
0.646
0.644
0.790
0.808
Cell
0.331
0.401
0.557
0.629
0.626
0.772
0.791
Exact
0.350
0.414
0.566
0.638
0.638
0.782
0.800
In table A column ? is the solution to the Laplace equation with essential bounboundary conditions defined by equation (b) and with a change of sign. Column ?
is the
result of evaluating (b) for the coordinates of the nodes, ? = tf + ? is the sum
of
the two and the problem solution. Note that here a domain integral problem has
been solved without the use of internal cells.
2.18 Other Approaches for the Treatment of Domain Integrals
141
(ii) The Dual Reciprocity Method
The Dual Reciprocity Method (DRM) was invented by Nardini and Brebbia [13,
14] in 1982 for the solution of elastodynamic problems. It is essentially a
generalized way of constructing particular solutions and it can be used to repre
sent
internal forces or body force distributions as well as for solving non-linear an
d
time-dependent problems. The method can be applied to define body force
distribution over the whole domain or only part [ 15].
It was realized early on that the approach could be used to solve a wide variety
of parabolic as well as hyperbolic time dependent problems [16]. A complete
description of how the method can be applied to vibrations has been presented by
Nardini and Brebbia in reference [ 17]. The case of transient analysis has been
studied in detail by Brebbia et al. [18-22] including the extension of the techtechnique to deal with non-linear diffusion problems and axisymmetric problems a
nd
a whole book has been published dealing with the method [23] and its applicaapplications.
The Dual Reciprocity Method starts by representing any given function b as a
combination of a series of known coordinate functions f where f= ??,, x) is a
function between a point ?y and any other point x, a* are unknown coefficients
associated with each of the/' functions. The /' are considered to originate at '
j'
different points, many of them on the boundary (see Figure 2.26). These points
are called 'poles' and the/' functions are the same type for all those points. T
he
form of b is hence as follows,
?
* = I Pf
B.191)
M is the number of poles equal to boundary nodes (N) plus internal nodes (L)
The next stage is to define the particular solution, corresponding to the generi
c
function fJ, i.e.
V2mj=/j B.192)
The uJ field can be found by integrating the above equation. One can then
compute the value of any associated variable such as q' by differentiating the
particular solution.
Equation B.180) can now be written as follows,
cV + J q*u dT.- J q*u* </? = - ? \a> f D2uJ))u* dill
B.193)
Each term on the right hand side of the above equation can be integrated by
parts resulting in only boundary integrals. (Notice that each 'j' term involves
a
142
Chapter 2 Potential Problems
particular solution localized at a 'j' pole). This gives
c'm' + f q*u dV - J q*u* dT
?
?
= I {ocj 0*6* + J <j*mj' ?/? - | u*QJ ?/?||
B.194)
This formula has only boundary integrals but the right hand side is not found
by using only one particular solution but by the addition of a series of terms,
each
of them representing the effect of a particular solution localized at a 'j' pole
.
After applying the usual boundary element discretization equation B.194) gives
rise to the following system,
HU - GQ = [HU - GQ] a B.195)
or simply
HU-GQ = Sa
B.196)
where
S = HU + GQ
B.197)
_ Boundary
Nodes
Figure 2.26 The total number of Poles given by the sum of all Boundary and
Internal Nodes
? and Q are matrices N X M each for the boundary solution. (M is the number
of points where the function ? has been applied or poles; N is the number of bou
ndary
unknowns as usual and L the number of internal nodes. Hence M = N + L). The
columns of these matrices represent the values of the it' and q' functions at th
e
different nodes for the case of the/* function acting at a particular 'j' pole.
Notice that the a coefficients are different from the values of the b function a
t
the points under consideration. Both are related through equation B.191), which
2.18 Other Approaches for the Treatment of Domain Integrals
143
expressed in matrix form gives a M X M matrix F which can be inverted, i.e.
B = Fa B.198)
where ? vector represents the magnitudes of the b functions at the poles. F is
a matrix whose elements are the values of/ at all points for each position 'j' o
f
the function/. Hence one can invert B.198) to find a vector, i.e.
B.199)
where E = F"'.
Variable
q
?
Results
Node
I
2
3
4
5
17
18
19
20
29
30
31
for Torsion of Ellipse for Different
X
2.0
1.706
1.179
0.598
0.0
1.5
1.2
0.6
0.0
0.9
0.3
0.0
Y
0.0
-0.522
-0.808
-0.954
-1.0
0.0
-0.35
-0.45
-0.45
0.0
0.0
0.0
/-
-0.680
-1.019
-1.357
-1.531
-1.587
0.349
0.418
0.574
0.646
0.643
0.789
0.807
/= +
-0.680
-1.020
-1.359
-1.532
-1.588
0.349
0.418
0.573
0.646
0.643
0.789
0.807
f functions
r
/= 1 or ?
-0.682
-1.024
-1.363
-1.536
-1.592
?.35?
0.418
0.573
0.646
0.643
0.789
0.807
Cell
-0.733
-1.046
-1.378
-1.549
-1.611
0.331
0.401
0.557
0.629
0.626
0.772
0.791
Exact
-0.8
-1.018
-1.322
-1.528
-1.6
0.350
0.414
0.566
0.638
0.638
0.782
0.800
2.18 Other Approaches for the Treatment of Domain Integrals
145
For the linear elements employed in the discretization, results using the three
different/functions are seen to be very similar, thus the use of/= 1 + r is reco
mrecommended as this is the simplest to apply, requiring no special consideration
. Note that
the cell collocation results are much less accurate. In table ? DRM results are
presented for the same problem with / = 1 + r considering different numbers of
internal nodes.
Table ? Results for Torsion Problem for Different Values of L
Variable
4
?
Node X Y
I 2.0 0.0
2 1.706 -0.522
3 1.179 -0.808
4 0.598 -0.954
5 0.0 -1.0
31 0.0 0.0
L= 17
-0.680
-1.020
-1.359
-1.532
-1.588
0.807
L= 13
-0.678
-1.017
-1.357
-1.530
-1.585
0.806
L = 9
-0.677
-1.016
-1.354
-1.525
-1.580
0.803
L = 5
-0.676
-1.013
-1.349
-1.517
-1.573
0.798
L= 1
-0.666
-0.995
-1.325
-1.499
-1.557
0.788
Exact
-0.8
-1.018
-1.322
-1.528
-1.6
0.800
It can be seen that in this case the sensitivity of the results to number of int
ernal
1 to L = 17 is approxinodes is small. The total variation of results from L
approximately 2%.
(iii) The Multiple Reciprocity Method (MRM)
The Multiple Reciprocity Method is a generalization of the Galcrkin technique
described in section 2.12 for taking domain integrals to the boundary. It has so
me
features which are similar to the DRM but instead of approximating the source
terms by a set of coordinate functions it uses a sequence of functions related t
o
the fundamental solution as will be shown. This sequence forms a set of higher
order fundamental solutions which permits the application of Green's identity to
each term of the sequence in succession. As a result the method can lead in the
limit to the exact boundary only formulation of the problems.
The MRM was presented in its present form by Nowak and Brebbia [ 12] [24]
who applied it to solve transient and Helmholtz type problems.
Consider again the case of Poisson's equation B.179) but now with a subscript
'o' on the right hand side term to differentiate from other similar functions wh
ich
will be generated during the solution, i.e.
V2u = b.
in ?
B.206)
146
Chapter 2 Potential Problems
where ? and ? are the potential and the source functions respectively.
The fundamental solution to Laplace's equation will also be called with a
subscript 'o' for similar reasons as explained above, i.e.
V2u* + A' = 0 B.207)
Applying reciprocity to the above relations one obtains the same expression as
before i.e. equation B.180), which is written again below for completeness
cV + f q*u dT + f bou* d?l = J </? * </?
B.208)
?
u
?
where g*= ??*/??.
The domain integral in B.208) can now be transformed into a series of equiequivalent boundary integrals. In order to do this one introduces a new function
uf
related to the fundamental solution ?* by the formula
V2m? = m*
B.209)
Thus the domain integral in B.208) can be expressed as follows,
f fc#M; Ml = J h,,(V2u*?
u
u
As the source function b0 is at present assumed to be a known function of
space, one can obtain the function V 2bt, analytically and hence one can define
a
new function b, such that
ht=V% B.211)
The domain integral on the right hand side of B.210) can then be written in a
form which is similar to the one for the previous domain integral and can be
expanded in the same way, i.e.
J b,T du = J {ft, M _
^'] dr + J u*(V%)dn B.212)
?
? ( cn on)
a
Notice that one can now compute a b2 function such that
b2 = V% B.213)
and continue carrying out this procedure as many times as desired.
The procedure can be generalized by introducing two sequences of functions
2.18 Other Approaches for the Treatment of Domain Integrals
147
defined by the following recurrence formula,
0,h2... B.214)
V72,.* _
V Ui+i Uj
The domain integral can then by expressed as
I bX </? = ? J \b} ? - uj+, &\ dT
B.215)
ii
j=o ? L vn
un J
Introducing B.21S) into the original boundary integral expression one obtains
[3] Brebbia, C. A. and Dominguez. J. Boundary Element Methods for Potential Prob
lems.
Applied Mathematical Modelling, I, 7, December 1977.
[4] Hess, J. L. and Smith, A. M. O. Calculation of Potential Flow about Arbitrar
y Bodies.
Progressin Aeronautical Sciences Vol.8. Ed. D. Kucliemann, Pcrgamon Press, Londo
n,
1967.
[5] Harrington, R. F., Pontoppidan, K., Abramhamsen. P. and Albertsen. N. C. Com
puComputation of Laplacian Potentials by an Equivalent-source Method. Proc. IEE. 1
16.
1715-1720 1969.
[6] Mautz, J. R. and Harrington, R. F. Computation of Rolationally Symmetric Lap
lacian
Potentials. Proc. IEE, M7. 850-852, 1970.
2.18 Other Approaches for the Treatment of Domain Integrals
149
[7] Jaswon, M. Integral Equation Methods in Potential Theory, I. Proc. Rov. Soc.
Ser.
A.. 275, 23 32. 1963.
[8] Symm, G. T. integral Equation Methods in Potential Theory, 11. Proc. Roy. So
c. Ser.
A., 275. 33-46. 1963.
[9] Jaswon, M. and Pontcr, A. R. A Integral Equation Solution of the Torsion Prob
lem.
hoc. Roy. Soc. Ser. ?., 273. 237-246. 1963.
[10] Brebbia.C. A. and Fcrranle, A. Computational Methods for the Solution of Ln
gineerina
Problems. Pentcch Press, London, Wiley. USA. 1986.
[II] Danson. D. J.. Brcbbia. ? A. and Adcy. R. A. BEASY. A Boundary Element Anal
ysis
System. Finite Element Systems Handbook. Springer-Verlag and Computational
Mechanics Publications, Southampton, 1982.
[121 Nowak, A. J. and Brebbia, C. A. The Multiple Reciprocity Method - A New
Approach for Transforming BEM Domain integrals to the Boundary. Eng. Analysis,
6C), 1989.
[ 13) Nardini, D. and Brebbia, C. A. A New Approach to Free Vibration Analysis u
sing
Boundary Elements, in Boundary Element Methods in Engineering (cd. C. A.
Brebbia), Springer-Verlag, Berlin and New York, 1982.
[ 14] Brebbia, ? A. and Nardini, D. Dynamic Analysis in Solid Mechanics by an Al
terAlternative Boundary Element Approach, Int. J. Soil Dynamics and Earthquake
Engineering, 2D), 228-233, 1983.
[15] Niku, S. M. and Brebbia, ? A. Dual Reciprocity Boundary Element Formulation
for Potential Problems with Arbitrarily Distribution Forces. Technical Note, Eng
.
Analysis, 5A), 1988.
[ 16] Nardini, D. and Brebbia, C. A. The Solution of Parabolic and Hyperbolic Pr
oblems
using an Alternative Boundary Element Formulation. In Proc. of VHth Int. Conf.
on BEM in Eng. Computational Mechanics Publications, Southampton and Boston,
Springer-Verlag, Berlin, 1985.
[ 17] Nardini, D. and Brebbia, ? A. Boundary Integral Formulations of Mass Matri
ces
for Dynamics Analysis. In Topics in Boundary Element Research, Vol. 2 (ed. C. A.
Brebbia) Springer-Verlag, Berlin, and New York, 1985.
[18] Wrobel, L. C, Brebbia, C. A. and Nardini. D. The Dual Reciprocity Boundary
Element Formulation for Transient Heat Conduction. In Proc. of the Vth Int. Conf
.
on FEM in Water Resources. Computational Mechanics Publications, Southampton
and Boston, Springer-Verlag, Berlin, 1986.
[19] Wrobel, L. ? and Brebbia, C. A. The Dual Reciprocity Boundary Element
Formulation for Non-Linear Diffusion Problems. Computer Methods in Applied
Mechanics and Engg. Vol. 65, pp. 147-164, 1987.
[20] Wrobel, L. C, Brebbia, C. A. and Nardini, D. Analysis of Transient Thermal
Problems in the BEASY system. In BETECH/86 (Eds. J. J. Connor and C. A.
Brebbia), Computational Mechanics Publications, Southampton and Boston, 1986.
[21] Brebbia, C. A. and Wrobel, L. C. Non-Linear Transinet Thermal Analysis usin
g
the Dual Reciprocity Method. In Boundary Element Techniques: Applications in
Stress Analysis and Heal Transfer, (Eds. C. A. Brebbia and W. Venturini), ComComputational Mechanics Publications, Southampton and Boston, 1987.
[22] Wrobel, L. C, Telles, J. ? F. and Brebbia, C. A. A Dual Reciprocity Boundar
y
Element Formulation for Axisymmetric Diffusion Problems. In Proc. of VHth Int.
Conf. on B.E.M. in Engg., Tokyo, 1986. Computational Mechanics Publications,
Southampton and Boston, Springer-Verlag, Berlin, 1986.
{23 J Partridge, P. Brebbia. C. A. and Wrobel, L. C. 'The Dual Reciprocity Bound
ary
Element Method'. Computational Mechanics Publications, Southampton and
Boston, and Elsevier, London, 1991.
[24] Nowak, A. J. and Brebbia, C. A. Solving Helmholtz Equation by Boundary
Elements using the Multiple Reciprocity Method. In Computers and Experiments
in Fluid Flow, (Eds G. Carlomagno and ? A. Brebbia) Computational Mechanics
Publications, Southampton and Boston Springer-Verlag, Berlin, ,1989.
150
Chapter 2 Potential Problems
Exercises
2.1.
Verify that the two-dimensional fundamental solution satisfies Laplace's
equation for
any point where * ? 0.
2.2.
The two-dimensional fundamental solution is ?* = A/2?I?A/?), Since it in
cludes a
logarithm of a non-dimensionless quantity its value depends on the scale. Will t
his
fact change the solution of a problem depending on its units?
2.3.
Using program POUNHH, solve the example of the figure with 12 linear ele
ments
and prescribing values of the potential on the whole boundary: linear variation
of ?
between nodes 1 and 4, ? = 0 belween nodes 4 and 5 linear variation of ? between
nodes 5 and 6, ? = 100 between nodes 6 and 8, linear variation of ? between node
s 8
and 10 and ? = 300 between nodes 10 and 1. Check that the results for the fluxes
at
the corners are very poor.
? mm
. . u-100
u-0
2.4.
Solve again the example of 2.3 prescribing values of the potential on th
e whole
boundary and introducing six new very small elements at the corners in such a wa
y that
the domain has blunt corners. Try several sizes of the corner elements to check
the
accuracy of the approximation.
2.5.
Using program POLFNBE, solve again the example of 2.3 with 12 linear ele
ments
and assuming that the values of the potential are known on the whole boundary.
Since the potential is known along two lines merging into each comer, determine
the
value of one potential derivative by differences along one of these lines and le
ave the
other as unknown.
2.6.
Derive, using the method of images, the fundamental solution for a semifinite region
with the condition that the potential on the free surface is zero.
2.7. The same as 2.6 but with the condition that the flux on the free surface is
zero.
Exercises
151
I and I
I
PxJ \fixj
which should be used with equation B.33) to compute internal fluxes.
2.9.
Write a subroutine (DERINPC) which given an internal point XP, YP, compu
tes
the integrals J I
I dV and f I ] </? along a constant element defined by i
ts
?'1*// i-A'>*//
extreme points.
2.10. Write a subroutine AFLUXPC) that using DERINPC and the boundary potentia
ls
and fluxes of (FI and DPI) of the program POCONBE computes fluxes at internal
points.
2.11. Assume an open channel section that extends from x = 0 to .v =
oo. At th
e extreme
x=0 there is u piston that moves harmonically with frequency w. When the
perturbations are small and the fluid is considered to be incompressible with ze
ro
viscosity, the equation that governs the motion is: V2/> = 0, p being the pressu
re, and
; "= />w2"n. where ? is the displacement along the n direction and p the density.
f'H
Nodes where
a relation
lietween p
<>
and may
<Vl
be written
It is known that I he solution that satisfies conditions at z = H (p = 0) and z
= 0
('. =0), is />(n,.x) = f AHcosK,zeK-* where ? = ^-^? (n= 1,2....). In order
Vj- / i
211
to solve a potential problem in a hounded region ? for a prescribed motion of th
e
piston, an artificial boundary with four constant elements is introduced at X = X
o.
Using as many terms of the series expansion of p(z, x) as nodes exist on the art
ificial
dp
boundary, determine a matrix that relates the nodal values of p and along that
on
boundary (Robin boundary condition).
2.12. Compute, using program POCONBE, the warping function of Saint Venant torsi
on
_U<1U2
2\dx3
C.11)
where f = 1,2. 3 and./ = 1, 2, 3. Another way of expressing C.11) is as follows,
where the comma indicates derivatives.
Sometimes the state of strain is defined using the strain components arranged
in a tensor (the strain tensor), i.e.
C.12)
r-si
where <:2, =12; =:,,; cJ2 = k23.
It is simpler to apply boundary conditions in terms of displacements rather
than in function of strains. Hence on ?, the following 'essentials' conditions c
an
be defined
or
)= 1,2,3
on ?,
on ?,
C.13)
where u} are the prescribed values. Note that the total ? surface of the boundar
y
is equal to ?, + ?2.
Constitutive Relationships
The states of stress and strains in a body arc related throughout the strain-str
ess
or constitutive equations for the material. For a linearly elastic material one
can
define two constants, called Lame's constants, ? and /< which are associated wit
h
J58
Chapter 3 Elastostatics
the volumetric and shear components. Then the stress-strain relationship can be
expressed as
Gjj = ).50?.?? + 2/ifilV
C.14)
where 8i} is the Kronecker delta (=1 for i=j and =0 for i^j). Notice that ckk
has only internal indexes and hence it implies a ratio of the three direct strai
n
components, and because of that is called the volumetric strain, i.e.
The inverse of C.14) can be written as
- akk + ffji C.16)
Cjj =
2/iCA + 2?) 2?
where akk = at, + <r,2 + ^??The Lame's constant can be expressed in terms of the more familiar shear
modulus G, Modulus of Elasticity ? and Poisson's ratio v by the following formul
ae,
? = G = E -; / ?
vE
C.17)
2A + v> (l+v)(l-2v)
The strain and stress in terms of E and ? can be written as
^--?*? + 1|-? C.18)
and
For some particular problems (specially in soil mechanics) one may prefer to
use the bulk modulus K.
In these cases one defines the deviatoric stress and strain components
<, = ",,->* A* C.20)
?b = <?, - ]?*A;
C.21)
Thus the constitutive equations are expressed as
<*;,= 2?4 p=-Kr,kk
C.22)
p is the mean pressure.
p=-J
C.23)
3.2. Basic Equations of Linear Elastostatics
159
and
? = / + fG = fi/CA - 2v))
C.24)
In general for isotropic elastic materials all material constants can be express
ed
in function of two independent constants.
The equations of equilibrium C), strain-displacement relations F) and consticonstitutive equations F) give a complete system of equations from which one can
determine the components of stress F), displacements C) and strains F).
Initial Stresses or Strains
In many problems one can have initial state of stress or strain due to temperatu
re
or other causes. Consider an initial strain for instance. In this case the elast
ic
strain components are obtained by subtracting from the total strain those initia
l
strains, i.e.
cfj = e}j-i?j C.25)
where v*ti indicates the clastic components, i:\j the total and ?y the initial s
trains.
One can now define the stresses using the elastic strains, i.e.
or
' =< + <*
C.27)
The tfj components arc called initial stresses and are defined as,
C.28)
It will be seen later on how the initial stress or strain components can be
analysed and included in the boundary integral formulation.
Notice that if the initial strains are due to temperature and the material is
thermally isotropic, one can write
4 = <xtt
C.29)
where a is the dilatation coefficient and 0 the difference in temperature. The v
alues
of a"j are given by
^
C.30)
160
Chapter 3 Elastostatics
3.3 Fundamental Solutions
The formulation of the boundary integral equations for elastostatics to be descr
ibed
in section 3.4 require the knowledge of the solution of the elastic problems wit
h
the same material properties as the body under consideration but corresponding
to an infinite domain loaded with a concentrated unit point load. This is the
fundamental solution of elastostatics and is due to Kelvin.
If the Equilibrium equations C.4) arc expressed in terms of displacements
components one obtains Navier's equations, i.e. consider C.4)
*<w + 6,-0
C.31)
Substitute above the stress strain relationships C.19), i.e.
? jl 5??? + ??
C.32)
and the strain displacement equation C.11)
The results are the Navicr equations or equilibrium equations in terms of
displacements, that is
1
-bi = O
C.34)
ft
Kelvin solution is obtained from equation C.34) when a unit contributed load
applied at a point 'i* in the direction of the unit vector <? i.e.
b, = A'e,
C.35)
An easy way of computing the fundamental solution is using the representation
of the displacement in terms of Galcrkin's vector. One assumes a vector G from
8?^A
v)
r
Notice that when Laplace's operator is applied to the Galerkin's vector given by
equation C.47) the computed value of F, differs from that of equation C.42) in a
constant, the resulting value being also the solution of C.39). Any value of G
different from that given by C.47) in r2 terms may also be used and in fact one
of those fundamental solutions of the biharmonic equation for two dimensions
was used in Chapter 2 (equation B.110)). These values produce fundamental
solution displacements u* identical to equation C.54) except for a rigid body
motion that is neglected because it does not change the solution of the problem
as
can be seen below.
3.3. Fundamental Solutions
163
Stresses at any internal point can be written using the strain-displacement
relutions C.11) and the strain-stress equation C.19). They can be expressed as,
C.55)
where the kernel Slkj has been obtained from ufk and will be written in full lat
er on.
The tractions or surface forces on the ? boundary with normal n can be written
through C.55) and equation C.9) as,
Pt = p?ke,
C.56)
where the traction components for three dimensional case are
7i [? ? ^'* + .'? + ( " 2v)(n'r* " ?*?.')| A57)
-v)r2\_dn
J
n, and nk are the direction cosines of the normal with respect to x, and xk. dr/
dn
is the derivative of the distance vector r with respect to the normal.
For 2 dimensional plane strain problems one has
4?A v)r\_0n J
Example 3.1
Consider for instance one component of the Galerkin vector, say G,, for the case
of (wo dimensional elasticity. Substituting it into equation C.36) will produce
the
following two displacement components (which correspond to a force acting in
the direction/= I), i.e.
(a)
_______
One can now substitute ihese components into the Navier equations C.34) which
are expanded into
/ I W, ???-] ?' .
\\-2vJ\dx\ dx2dxtj ti
(b)
164
Chapter 3 Elastostatics
Surface ?
general point on
the surface
Point i of
application of
load
I.
(a) Geometric definition
x2.u2
Unit load in xj direction
X3
x,,u,
(b) Displacement components of the
fundamental solution of the surface
a
tj
rt
This expression is a generalized statement that can be used to obtain the bounda
ry
integral equations. Having established this starting principle one can now retur
n
to expression C.65) and use as weighting functions the fundamental solution
presented in 3.3, which was obtained for a point load b, = &! along the directio
n
of the unit vector eh i.e.
??? + ?!?, = 0 C.67)
The fundamental solution may be written as before, i.e.
M* = H(*f,
P* =
C.68)
where hJ, pfk are ? components of displacements and tractions due to a unit poin
t
load in the / direction. The first integral in C.65) for a particular direction
e, of
the unit load becomes
J atjjtk ilu = J afjji, tin = - J A'(t'( JQ = -u\e,
C.69)
where u\ represents the / component of the displacement at the point / of applic
ation
of the load.
Equation C.65) can now be written to represent the three separate components
of the displacement at i by taking the three directions of the point load at '/'
3.4. Boundary Integral Formulation
167
Point on the surface
Figure 3.3 Interpretation of the components of the distance vector r
independently, i.e.
11
?2
C.70)
Notice that when one applies a unit point load along a particular direction */',
the tractions and displacements at any point in the domain have components
along the three or two directions (equations C.68), and C.51) to C.58)) while
terms of the type alti only are different from zero along the direction of the l
oad.
This leads to the fact that for a given direction T at the point the first term
of
C.65) only produces displacements along the direction of the load (first term of
C.70)). The rest of the terms however include products for all the components.
Equation C.70) can be written in a more compact way if one considers the two
parts of the boundary together (i.e. ? = ?, + ?2) and applies the boundary
168
Chapter 3 Elastostutics
conditions at a later stage. In this case C.70) becomes
i + J Pikuk dr = i ufkpk dV+l ufkbk du C.71)
?
?
u
This equation is known as Somigliana's identity and gives the value of the
displacements at any internal points in terms of the boundary values uk and pt,
the forces throughout the domain and the known fundamental solution. Equation
C.71) is valid for any particular point V where the forces arc applied.
Boundary Points
Somigliana's identity gives the displacement at any internal point once uk and p
k
arc known at every boundary point and consequently only when the boundary
value problem has been solved the values at the internal points can be calculate
d.
However, since equation C.71) is valid for every point in Q including ?,? bounda
ry
integral expression can be obtained by taking C.71) to the boundary. This
expression is applied at different points on the boundary to produce a system of
equations which once solved give the boundary values.
When 7' is taken to the boundary, however, the integrals have a singularity
and we need to analyse this behaviour in the same way as it was done in Chapter
2, section 2.2 for potential problems. If we consider that the boundary is smoot
h
at T one can supplement it by a hemisphere (figure 3.4) with centre at '/' and a
small radius e which will afterwards be taken to the limit, i.c.c -* 0.
There are two types of boundary integrals in equation C.71). Consider first
the one on the right hand side and write it in function of ?? surface of the hem
isphere,
i.e.
J ufkpk dT ~ lim < J ufkpk dT > + lim < J wj/>k dT >
C.72)
?
t-o (.r-r,
J *-*o (r,
J
The first integral on the right hand side of C.72) will simply become an integra
l
on the whole boundary ? when 6 -> 0. The second integral can be written as,
f )
pllimtjufbdr} C.73)
^o (.rt J
Noticing that the fundamental solution is of order I/e and the surface integral
in
C.73) will produce a e2, one can conclude that C.73) will tend to zero as <; ->
0, i.e.
J?7
C.74)
-? (r, J
In other words the integral investigated is not affected by the singularity at 7
\
3.4. Boundary Integral Formulation
169
The left hand side integral in C.71) however behaves differently. If one writes
it as
J p?kuk <1? = lim I J tttMrl + lim-hpJMri
C.75)
I
?-0 (r-r,
J *-0 lr,
J
one can sec that the limit of the last integral can be written as
C.76)
The pfk values are now of l/v.2 order while the terms resulting from integration
over the surface of the hemisphere are of order r.2. Hence the integral C.76) do
es
not vanish when r. -0 but produces a free term. By substituting the values of pj?
as given in equation C.57) and integrating over ?? one finds
-i5 C.77)
Hence the left hand side integral C.75) can be written in the limit as
J rfkuk dr - i^ul = f pfkuk dT - X
C.78)
where the integral on ? is defined in the sense of the Cauchy Principal value. T
his
is demonstrated in detail in example 3.2 where the different terms of the
fundamental solution arc integrated one by one.
Therefore for boundary points equation C.71) transforms into
<U + J P>k dr = f ufkPk dr + J MfA dn C.79)
where the integrals are in the sense of Cauchy principal value and where ? is sm
ooth
at'/', c\k = }(\?. When '/' is at a point where the boundary is not smooth, the
value
of the integrals in equation C.78) give different results and it is generally di
fficult
to obtain a general expression in three dimensions.
Fortunately, explicit calculations of this value is not usually necessary as it
can be obtained using rigid body motions as will be shown in section 3.5.
Boundary equation C.79) permits to solve the general boundary value problem
of claslostatics. If displacements arc known over the whole domain, equation C.7
9)
produces an integral equation of the first kind, if tractions are known over ail
the
boundary an integral equation of the second kind is obtained and finally a
combination of both types of boundary conditions results in a mixed integral
equation.
170
Chapter 3 Elastostatics
Kxample 3.2
Consider the behaviour of the two types of integrals in C.71) for the case of th
e
point T being on the smooth boundary ? surrounded by a hemisphere as shown
in figure 3.4. In the limit the radius of the hemisphere will tend to zero.
The first type of integral is as follows.
I *ft* ?? = lim \ I ukplkdr\ + \imUukPtkdr\
(a)
?
?-0 lr-?
J i-0 U\
J
Consider the ?? integral only, i.e.
-o (.r,
r.
k
?^^]]Lbr
(b)
Note that r. = r. Consider figure 3.5 where for simplicity a spherical system of
coordinates is used. For this particular case the second term in equation C.57)
will disappear as,
dr dr dr dr dr dr
dx, dxk oxt vxk oxk dxt
Hence one only needs to consider the first term in the integral, that is.
Note that PrjOn = 1.
This can be expanded taking into account the geometric relationships shown
in figure 3.5, for instance when / = 1
/ = lim < f {u'i(l '.
c-0 I f,
+ 3u'ielei)
}
(e)
8??A - v) J
The et are unit vectors in the x, direction (see figure 3.5) such that
dr r,
e, = n, - - - ??, ?
3.4. Boundary Integral Formulation
171
Surface ?,
All other surfaces
are ?-??
Figure 3.4 Full boundary surface ?? assumed hemispherical for integration
purposes
r cos II
Figure 3.5 Geometry definitions
After integration we find that some of the integrals are zero and the Final resu
lt is.
[0 8A-v)
(f
172
Chapter 3 Elastostatics
The same can be shown to apply for / = 2 and / = 3. This result can then be writ
ten
as,
{
(g)
The integral
can also be written
J p,MS,dr+jpku1kdr
(i)
r-r
rt
but it can be easily shown that
Km J ???? i/? = 0
(j)
and therefore this integral does not produce any new term.
Hence the final expression for the integral on a smooth boundary is,
il + J "*/>* ?? = | pku*k dT + f Vf* <?
(?)
?
?
?
3.5 Boundary Element Formulation
In order to solve the integral equation numerically, the boundary will be discrc
tized
into a series of elements over which displacements and tractions arc written in
terms of their values at a series of nodal points. Writing the discrctized form
of
C.79) for every nodal point, a system of linear algebraic equations is obtained.
Once the boundary conditions are applied the system can be solved to obtain all
the unknown values and consequently an approximate solution to the boundary
values problem is obtained.
It is now more convenient to work with matrices rather than carry on with
the indicia! notation. To this effect we can start by denning the ? and /> funct
ions
which apply over each element '/, i.e.
C.80)
and
? = ??':
C.81)
where u^ and pJ are the element nodal displacements and tractions, of dimensions
3 x ? for three dimensions and 2 x Q for two dimensions, Q being the number of
3.5. Boundary Element Formulation
173
nodes on the element, u and p are the displacements and tractions at any point
on the boundary ?,, i.e.
P =
(??
?2
u =
C.82)
The interpolation function matrix ? is a 3 x 3Q (or 2 x 2Q two dimensions) array
of shape functions, i.e.
?=
? ?
? ?
<t>2
? ?? ? ? ?2 ?
.0 0 ?? 0 0 ?2
??1
?? 0 0
? ?? ?
? ?
C.83)
These functions are the standard two dimensional finite clement type function
discussed in section 2.7 (sec Figure 3.6).
Notice that the body forces at any point on the 12 domain can also be expressed
in vector form in function of the three components, i.e.
b =
The fundamental solution coefficients can be expressed as,
C.84)
P* =
??? Pi 2 P*3
P*l P*2 P*J
L P*. P*2 P*3 j
*2 J3
matrix whose coefficients, pfk,
are the tractions in ? direction
due to a unit force at V acting
in the T direction
C.85)
u' =
matrix whose coefficients ufk are
the displacements in the 'fc' direction C.86)
due to a unit force at T acting on
the '/' dirccton
1 22 23
-1 32 33With this notation equation C.79) valid for each; point can be rewritten as foll
ows,
cV + | p*u A? = J u*p ?/? + | u*b (/Q C.87)
where c' = { fr smooth boundary. Otherwise it will be a 3 x 3 (or 2 x 2 in two
dimensions) array.
174
Chapter 3 Elastostatics
Nodes
Boundary
elements
x2
(a)
Boundary
elements
Nodes
1
Elements
Nodes
*2
(c)
Figure 3.6 Three-dimensional body divided into (a) constant boundary elements,
(b) linear boundary elements and (c) quadratic boundary elements
3.5. Boundary Element Formulation
175
Notice that the Cartesian coordinates of the boundary may also be written in
terms of nodal coordinates to define curved elements. If this is the case we wil
l
need to transform from one to another system and this transformation will involv
e
introducing a Jacobian as it was shown in section 2.7. We will discuss this in w
hat
follows.
Consider now that we substitute the above functions into equation C.87) and
discretize the boundary obtaining the following equation for a nodal point.
c'u'+ X i f ?*?</???'= ? < J u*(t>dr}pJ+ ? -M a*bdn\
C.88)
/=' li'j
J .Hi In
J s=i (h.
j
Note that summation for j = I to NE indicates summation over all the NE elements
on the surface and ?{ is the surface of a '/ element, a1 and p1 are the nodal
7=1
/=l
s=l
where N is the number of nodes, u' and pj arc the displacements and tractions at
node 7'. The influence matrices H and G (now 3 x 3 for three dimensions and
2 x 2 in two dimensional cases) arc
C.95)
I ?,
where the summation extends to all the elements to which node '/ belongs and q
is number of order of the node '/ within element t. For constant elements, the
summation extends only to one clement, t=j and ? is the identity matrix.
B" = | u*b du
a,
Calling
Hy = fty
ifi/j
C.96)
if i =j
cquation C.94) for node T becomes,
The contribution for all T nodes can be written together in matrix form to
give the global system equations, i.e.
HU = GP+B
C.98)
Notice that the elements c1 will be a series of 3 x 3 submatriccs on the diagona
l
H (or 2 x 2 in two dimensional cases). The elements of these submatriccs are not
simply given by the solid angle but can become very cumbersome to compute
178
Chapter 3 Elastostatics
analytically. Fortunately this is not required as they can be found by considera
tion
of rigid body movement as we will see shortly.
The vectors U and P represent all the values of displacements and tractions
before applying boundary conditions. These conditions can be introduced by
rearranging the columns in H and G, passing all unknowns to a vector X on the
left hand side. This gives the final system of equations, i.e.
AX = F C.99)
Notice that the ? vector has been incorporated in F. Solving the above system
all boundary values are fully determined.
Rigid Body Considerations
As it was pointed out the diagonal submatrices H" in H include terms in ?" and
c1. Difficulties appear when trying to compute explicitly these terms particular
ly
at corners due to the singularity of the fundamental solution. Assuming a rigid
body displacement in the direction of one of the cartesian coordinates the tract
ion
and body force vector must be zero and hence from C.98)
HI = 0 C.100)
where I* is a vector that for all nodes has unit displacement along the V direct
ion
(</= 1, 2 or 3) and zero displacement in any other direction. Since C.100) has t
o
be satisfied for any rigid body displacement one can write,
N
H" = ~ I H'v (forj ? i) C.!01)
which gives the diagonal submatrices in terms of the rest of the terms of the H
matrix.
The above considerations arc strictly valid for closed domains. When dealing
with infinite or semi-infinite regions equation C.101) must be modified. If the
rigid
body displacement is prescribed for a boundless domain the integral
\
C.102)
over the external boundary ?, at infinity will not be zero and since the tractio
ns
p* are due to a point load, this integral must be,
Jp*</r=-I
C.103)
3.S. Boundary Element Formulation
179
where I is the 3 x 3 (or 2 x 2 in two dimensional
cases) identity matrix.
The
diagonal submatriccs for this case are,
H" = I - ? H" (for )? i)
C.104)
Internal Points
Somigliana's identity C.71) gives the displacement at any internal point in term
s
of the boundary displacements and tractions. Considering again its integral
representation as in C.88) one has,
?? (
")
?? ?
")
C.105)
=i in. J
where ?} is once more the surface corresponding to element j and V is now an
internal point. The internal point displacements in terms of the nodal displacem
ents
and tractions can be written in the same way as C.94), i.e.
n'= t C'V- t fl""'+ I B"
C.106)
The terms GIJ and H'7 consist of integrals over the elements to which node j
belongs. Those integrals do not contain any singularity and can be easily comput
ed
using numerical integration. Terms like B's however will contain a singularity
(notice that they are domain terms and the point /' is now in the domain) and
special care should be taken when computing them numerically. Being domain
integrals however their order of singularity is one less than the integrals on t
he
boundary and consequently can be more accurately computed using numerical
integration formulae.
For an isotropic medium the internal stresses can be computed by differentiating
the displacements at internal points and introducing the corresponding strains i
nto
the stress-strain relationships, i.e.
C.107)
a = _?^L Sij ?Ul + J-"' + ?
v I 2v f'.x, \dxj dxj
180
Chapter 3 Elastoslatics
After carrying out the derivatives inside the integral equations one obtains,
?.,,,
All derivatives are taken at the internal point under consideration, whichjs the
point of application of the fundamental solution. Taking the corresponding
derivatives of the fundamental solution the above equation can be written, in a
compact form, as,
m
C.109)
where the third order tensor components DklJ and SUJ arc
Dm - - {{1 - 2v){dkirj + V.i ~ Va)
4??A - v)
r,r.h)
1
C40)
+ A - 2v)(Pnkrjrj + iijSik + nfijt) - A - 4r)nt?ui
4?A - v)
C.111)
The above formulae are applicable for 2 or 3 dimensions. For the former case
a = 1, p = 2 and ? = 4 and for the latter a = 2, /? = 3 and ? = 5.
All the derivatives indicated by commas arc taken at the boundary point xf
of figure 3.8, i.e.
This derivative is equal and opposite in sign to those taken at an internal poin
t.
Equation C.109) is discretized by dividing the ? boundary into a summation
over all the element surfaces and assuming the corresponding interpolation
functions for uk and pk.
3.5. Boundary Element Formulation
181
Boundary point
-> x,
! i^urc 3.8 Definition of derivatives required in lite internal stress formulae C
.109)
The values obtained for the internal stresses using the above formulae arc in
general more accurate than those computed using other numerical methods and
similar discretization. The same can be said of the internal displacements compu
ted
through C.105). However when the internal point is very close to the boundary
(say less than 1/4 of the smallest length of the nearest clement) because of the
peak in the fundamental solution, special numerical integration schemes have to
be used to obtain accurate stresses and displacements.
Notice also that the values of displacements at the boundary arc known from
the integral equation solution. The same can not be said of the stresses since t
he
stress vector has more independent terms than the traction vector. The problem
will be studied in the next section.
Stresses on the Boundary
The numerical solution of the governing equations produces all the boundary
displacements and tractions. In many practical applications however the boundary
stresses rather than the tractions arc required. One possibility would be to tak
e
C.109) to the boundary but this produces higher order singularity of a type that
has not yet been effectively resolved. For three dimensional problems, the Dkij
and
Skij terms contain singularities of order \/r2 and I/*'3 respectively and taking
equation C.109) to the boundary requires special consideration of how the princi
pal
values should be computed.
182
Chapter 3 Elustostatics
The simplest way of determining (he stress tensor at boundary points is to compu
te
its components from the known boundary tractions and displacements. Let us
assume a three dimensional case and a local cartesian system of coordinates at t
he
boundary points where the stresses are to be computed (see figure 3.9). It is ea
sy
to see that,
C.113)
a 33 = P'i
where the dash indicates local coordinates. In addition to the tractions a discr
ete
expansion for the boundary displacements over the element is also known as given
by C.80), i.e.
u' = R'<t>uJ
C.114)
where R is a transformation matrix from global to local system. Four components
small and it is limited by the numerical problems that may be originated by the
existence of two sets of equations whose coefficients arc very close to each oth
er.
In practice excellent results are obtained if the distances arc not too small. W
hen
the corner node is duplicated a small gap may be left between the two nodes
(figure 3.11 (a)) or a small element may be assumed between the two (figure 3.1
Kb)).
In the latter case, tractions over the small elements arc assumed to be />',, p'
2 and
p\, p for nodes 2' and 2" respectively.
Another version of the double node approach is the use of discontinuous
elements already presented for potential problems in section 2.7. This consists
of
displacing inside the element the nodes that meet or that would meet at corners
or edges (figure 3.1 l(c)). The approach is very simple and effective and has th
e
added advantage that it can model better corners with high stress concentrations
.
When used to model singularities - such as in fracture mechanics problems the results using discontinuous elements have converged well to the correct solu
solution. The discontinuous elements formulation for two-dimensional linear elem
ents
will be presented in section 4.6.
The problem of having more than two unknowns at a comer can also be
solved as follows. If the tractions at both sides of a corner arc unknown the di
splacedisplacements along the two elements to which the node belongs will lie known. I
n such
a case the displacement derivatives along those elements can be obtained by
derivation of the shape functions. Since the two elements follow different direc
tions
all the components of the strain tensor at the corner can be written in terms of
nodal displacements of those elements and the stress tensor ami :iny traction at
the
3.6. Treatment ?? Domain Integrals and Body Force Terms
185
(a) Corner with gap
(b) Corner with small element (c) Corner with discontinuous elements
Figure 3.11 Corner tractions and displacements modelling
corner may be computed. This procedure can not be applied for corners where a
singularity of the stress tensor occurs.
3.6 Treatment of Domain Integrals and Body Force Terms
Domain integrals are important in boundary elements as they can not only
represent body or thermal forces but also a wide range of non-linear effects.
Initial Stress
Many problems produce an initial stress or strain field and in this case one nee
ds
to differentiate- between the clastic, initial and total components of those var
iables,
if an initial stress stale exists, the clastic stresses arc given by (sec sectio
n 3.2)
C.120)
represents the 'total' stress and <?? the initial one.
186
Chapter 3 Elastostatics
If we rewrite the starting equation C.66), taking into consideration fundamental
solution unit forces acting in the '/' direction, one can write
u
rj
r,
+ J (" *-*>??<*?
C.122)
11
Substituting equation C.120) one can write,
J hkufk dil - J a)kv.fjk dil - \ *?,& dil
11
11
11
= - J ftU <*? - J ftiij </? + J (? - uk)pfk dV C.123)
??
?,
Ti
Integrating by parts again one obtains
dil + 1 (J^jUk dil - J <7?** '/Q
u
ii
= ~ I /VU ^?- J Pkufk dr + J ???* ?/? + J ?? <T C.124)
Ij
?|
?,
?>
Notice that the integration by parts has been carried out on a\k and not on
ajlt because the former is the one related to the total displacements. Equation
C.124) gives in more compact form, and after substituting for the fundamental
solution, the following expression
J iSp rfr + J V,t'/ - J < <ln = ' + f /S* '/r
P.I25J
?
ii
it
?
This demonstrates that the initial stress (and similarly for strain) field can b
e
treated in a similar way as a body force field, hh, although it is generally dif
ficult
to transform the a term into a boundary integral.
Body Forces
Let us study in this section how the body force term in C.125) can be taken to
the boundary. The integral under study is
\hkufkdu
C.126)
12
The integral does not include any unknown values but in order to compute it
numerically the whole domain has to be discretized into cells which represents a
3.6. Treatment of Domain Integrals and Body Force Terms 187
considerable difficulty and diminishes the elegance and computational efficiency
of (he method. Fortunately domain integrals can be avoided in many practical
cases by reducing them to the boundary. Consider in this case that the body forc
e
components bk can be obtained from a potential function ? such that,
tlxk
C-127)
where the potential ? is assumed to satisfy the following harmonic relationship
?2? = ?0 = constant
C.128)
Then integral C.126) can be rewritten as
vxk/
8
??*
= ? 0??</?-?'*<??
C.129)
u t'xk a dxk
where derivatives by parts have been used. For the divergence theorem one can
now write
du*
I ufkhk <? = J ?*??? dT - J 7"* du
C.130)
?
i
a oxk
where nk is the direction cosine of the normal to ? with respect to xk axis. In
order
to take the last integral to the boundary one can write the fundamental solution
e
a. is the coefficient of thermal expansion. Hence the thermoelastic problem is a
particular case of the clastoslatic problem with body forces.
The boundary integral equation for a thermoelastic body without any other
types of body forces is
c,A + [ ????* dr = { ufkpk dT +1 ufk(y0)nk dT
?
Comparing lo C.139) we can write the potential of the equivalent body forces as,
?^-?? C.144)
Taking into account that for steady stale conduction one has V2fl = 0 the
domain integral in C.143) then becomes a boundary integral and C.143) can be
written as
C.145)
i + J /'?* dr = J ufkPk dr + \ -'{ J (Glk.k/h,j - ???0?
?
?
2( i ) ?
190
Chapter 3 Elastostatics
The above integral equation may be discrcti/cd in the usual way. ?? Ihe
functions in the last integral of equation C.145) are known and may be easily
integrated by numerical procedures over the boundary elements on the surface.
Thermoelastie problems may also be studied considering the temperature effects
as initial strains, i.e
j
j
C.146)
with the initial stresses given by
o% = - U4<),.k + 2,?) C.147)
The values of u"jk are
\
C.148)
- -\lNjk = -????
Substituting equation C.148) into the initial stress term of the elastic equatio
n
C.125) one obtains,
- J ??.?)? <Kl = J 70Sjkr.rjk <Kl f yOufa Ml
u
?
u
= - f 70 *w <lu + I ?0"8?* '/? C.149)
?
When equation C.149) is substituted into C.125) without any other type of body
force, one obtains once more equation C.143) that after taking the domain integr
al
to the boundary gives C.145). As could be expected both interpretations of Iherm
oclastic effects give identical results.
?.7 Subregions in Elasticity
In the previous sections the boundary elements for homogeneous linear elastic
isotropic media have been formulated. In many cases media that are not homogeneo
us
but consist of several zones each one being homogeneous must be analysed. Those
problems may be studied using the above formulation of the BEM and continuity
conditions. All the boundaries of the body have to be discretized, including int
ernal
boundaries that separate homogeneous zones within the medium. The equations
formulated for every homogeneous zone plus the displacement and traction
continuity conditions over the internal boundaries produce a system that may be
solved once the external boundary conditions are taken into account.
Consider the problem shown in figure 3.12 consisting of three zones of different
elastic materials. A two-dimensional domain has been represented for simplicity;
however, the procedure described below applies to both two and three-dimensional
problems.
3.7. Subregions in Elasticity
191
Figure 3.12 Zoned media
The following symbols are used:
?1 : external part of the boundary of zone Q1
?" : boundary between zones ft1 and QJ
u', p1 : nodal displacements and tractions at nodes on the boundary ?' of
zone ?'
u", pIJ : nodal displacements and tractions at nodes on ?1' as part of zone Cl'
H', G' : parts of the H and G matrices obtained for zone Q1 that multiply u1
and p1, respectively
H'J, G'J: parts of the H and G matrices obtained for zone Q' that multiply u'7 a
nd
pu, respectively.
The BEM equations for the three homogeneous zones of figure 3.12 are
[H1
HI3]{u12
.?13
:[Gl G12 G13] p12
for zone Q1
H21 H2I=G21 p21
for zone fi2 and
C.150)
C.151)
C.152)
for zone ?3. The traction equilibrium conditions and displacement compatibility
192
Chapter 3 Elastostatics
conditions over the internal boundaries ?'7 arc
.12 .
.21
C.153)
that transform equation C.151) and C.152) into
C1 G21]{p"} =
[H3 H31 G31]
,13
C.154)
C.155)
The last two equations plus equation C.150) can be rearranged into
,i
-Hl
0
. 0
0
0
H3
H12
H21
0
-G12
G21
0
13
0
??
-G13"
0
G31
U
12
13
G1 0
0 0
L 0 G3
C.156)
The above system of equations may be solved once the boundary conditions on
?1 and ?3 are prescribed. The total number of unknowns is equal to the number
of nodal degrees of freedom over the external boundaries plus twice the number
of nodal degrees of freedom over the internal boundaries.
The subdivision of the region into several zones may be used also in
homogeneous media as a way of avoiding numerical problems or improving
computational efficiency. For instance problems that include cracks or notches,
as the one shown in figure 3.13, present numerical difficulties when the boundar
y
is discretized due to the proximity of some of the nodes. The difficulties disap
pear
if the region is divided into two zones in a way that the nodes thai arc very cl
ose
belong to different regions. Another situation where the subdivision of a homohomogeneous region may be useful corresponds to problems with a large number of
unknowns, as the one shown in figure 3.14. in those cases the subdivision transtransforms the fully populated matrices into banded matrices, which arc more con
venient
from a computational point of view. In those cases, the increase in the number
of unknowns because of the internal boundaries must be small, otherwise the
subdivision will be worthless.
3.8. Indirect Formulations
193
Figure 3.13 Subdivision to avoid numerical difficulties in a problem with a notc
h
Figure 3.14 Subdivision to increase computational efficiency
3.8 Indirect Formulations
The fundamental boundary integral equation used throughout this chapter is
known as the direct integral representation and gives displacements at internal
and boundary points in terms of boundary tractions and boundary displacements.
There arc some other integral representations where the displacements are writte
n
in function of variables which arc not explicitly the boundary displacements or
tractions. Those representations arc the basis of the so-called indirect boundar
y
clement formulations.
Let us consider a state ait, c,y. Up pj over the domain ? of figure 3.l5(a). We
can now define another state over the complementary domain Q' (figure 3.15(b))
with variables a'ih i:]j, u'(, p). The tractions over the external region of Q'
will be
referred to the normal ? of the internal domain as shown in the figure.
It is easy to establish a reciprocity relationship for the complementary domain
between the fundamental solution applied at a point in ft and the complementary
state indicated by primes. This gives.
C.157)
194
Chapter 3 Elastostatics
This relationship can be subtracted from the integral equation for the reference
state, as given by C.71), i.e.
C.158)
Notice that the body forces are not included for simplicity.
this type of problem. The computer time per integration over a boundary clement
Figure 3,16 Rigid foundation embedded in the soil
3.8. Indirect Formulations
197
/ / /
Kigurc 3.17 Discrctizalion for one quarter of a square embedded foundation
450
oo
Figure 3.18 Effect of the interface discretization on the foundation stiffness.
(Results arc normalized as follows: Horizontal stiffness: K,n()rm = K,B-v)/GB;
Vertical stiffness: Kznorm = Kz(l - v)/GB; Rocking stiffness: Kyynorm = Kyv(l v)/
GB3; and Torsional stiffness: KKnorm = K,7J/GB'.)
198
Chapter 3 Elastostatics
50 0
400
100
Single Sol.
Double Sol.
00
02
08
10
04
06
B/A + B)
Figure 3.19 Effect of the free surface discretization on the foundation stiffnes
ses
grows however due to the greater amount of terms involved in the fundamental
solution. Another computationally more efficient approach is based on the use of
Kelvin's solution in conjunction with another point load solution following the
same direction but located at the image point of the first with respect to the s
oil
surface. The use of this solution reduces to zero the surface tractions with the
exception of pl3, p23 and piu pi2 which are expected to have a small effect in
soils. Figure 3.19 shows how when the double solution is used the result becomes
,
almost independent of the amount of free surface discretized and the free surfac
e
does not have to be considered.
Many other soil problems can be studied using (he boundary clement method.
Some applications in zoned media are shown in figure 3.20, where it can be seen
that the soil can be inhomogeneous.
3.9 Axisymmetric Problems
There are many elastic problems that present an axisymmetric geometry and very
frequently also axisymmetric loading conditions. In the following, it is first
explained how boundary elements may be applied to axisymmetric problems with
respect to both the geometry and the loading by taking advantage of the symmetry
that reduces a three-dimensional analysis to two uncoupled plane domain
problems; one with two degrees of freedom per point (radial ami axial) and anoth
er
with one degree of freedom per point (tangential). Non-axisymmetric loads are
studied later. The first boundary elements formulation for axisymmetric elastic
problems was published in 1975 [II], [12]. It was based on the Somigliana's
identity obtained from the application of the reciprocity theorem between the ac
tual
/1 = 0
? 0
,-.
?.COS II
(.*
COS ?
0 +
nO
0 +
;,
+ ?
<-.
sin
v
sin
Htf)
?0)
3.10. Anisotropic Elasticity
203
C.174)
where V indicates the symmetric terms and '' the antisymmetric ones.
For each Fourier mode amplitude a discretized boundary equation tike
equation C.171) may be written with the only difference being that p* and u* are
now weighted by a sine or a cosine function and the integrals around the axis ar
e
of the form
"jf u* sin nO ilO; "\ af cos nO dO
C.175)
?
1)
It is worth noting that since sin nO has zero value at 0 = 0, the i point cannot
0 to compute the amplitude of those terms of the Fourier series
be located at 0'
.
One only has to move i to a point, for instance 0' =
jr/2n,where the amplitude
is not zero. This change is easily taken into account by a shift of the origin o
f 0
in equation C.170).
3.10 Anisotropic Elasticity
The constitutive relationships described in section 3.2 are valid for isotropic
cases
for which the behaviour of the material can be described in terms of only two
constants. In more general cases the material can be orthotropic or generaly
anisotropic. For a three dimensional anisotropic case the number of different cl
astic
constants is 21 and they can be expressed in matrix form as,
o = De C.176)
where D is a 6 x 6 matrix. The a and ? are the usual six stress and strain
components. The above elastic relationship can also be written in index form
which gives a notation more consistent with the previous sections, i.e.
au = dlikfiu
C.177)
(lijkl are called the rigidity coefficients. The inverse of C.176) produces the
elastic
compliance, cljkl, i.e.
hj = Vw
C.178)
Although there are in general 21 constants for an elastic material, their number
is reduced when the material structure has one or more planes of symmetry.
204
Chapter 3 Elastostatics
Materials with the orthogonal planes of symmetry are said to be orthotropic and
the number of independent coefficients reduces to 9.
in section 3.6. The other term can be integrated on the domain using cells or
transformed into a 'body force' type term and then taken to the boundary. Let
us consider the term on its own, i.e.
dU
C.187)
Next one integrate by parts C.187) which gives
-SdkjJufkdn + Spkutkdr C.188)
n
?
The first integral can now be interpreted in function of a fictitious body force
such
that
?? = ?? C.189)
206
Chapter 3 Elastostatics
Hence equation C.188) and C.189) can now be substituted into C.186) to give
+ J ?%? dT = J S( A + ?) /? + J ?,? + ?) <?1 C.190)
pk and St are acting now as out of balance forces which represents the anisotrop
ic
effect. The problem can be solved iteratively by finding the first solution with
[>k
and ????, and then computing their values and resolving the system as many
times as required. The influence matrices H and G are always the same. In matrix
form equation C.190) can be written as,
HU = CP + GP + DB + DB C.191)
The equations at the beginning of the iteration process are simply,
HU = GP + DB
C.192)
and from then on one solves the following equation
DB + B' C.193)
where B' = GP + DB, where P, ? terms result from the previous step.
While the hk terms may be reduced to the boundary as shown earlier the main
problem remains how to convert Bk terms into boundary integrals. If this is not
possible one will need to divide the domain into cells as explained in section 3
.5.
A general way of reducing body force terms to the boundary is by using
particular solutions. This has been generalized in references [16] to [21] by us
ing
a technique called the dual reciprocity method (DRM). Although this is an
interesting development with many applications in boundary elements it falls
beyond the scope of this book and for the further information the reader is refe
rred
to the work of Brebbia and his collaborators Nardini, Wrobel and Tang given in
the above articles.
References
[1] Alarcon, E., Brebbia, C. A. and Dominguez, J. The Boundary Element Method in
Elasticity, Intern. Jnl. of Mechanical Sciences, 20, 625-639, 1978.
[2] Rizzo, F. J. An Integral Equation Approach to Boundary Value Problems of Cla
ssical
Elastostatics, Quarterly of Applied Mathematics, 25, 83-95, 1967.
[3] Love, A. E. H. A Treatise on the Mathematical Theory of Elasticity, l5bver,
New
York, 1944.
[4] Cruse, T. A. and Rizzo, F. J. A Direct Fomulation and Numerical Solution of
the
General Transient Elastodynamie Problem I, Journal of Math. Analysis ami Applica
Applications. 22, 244, 1968.
[5] Cruse, T. A. A Direct Formulation and Numerical Solution of the General Tran
sient
Elastodynamie Problem II, Journal of Math. Analysis ami Applications, 22,341,196
8.
Exercises
207
[6] Rrchhiii, ?. ?. The Boundary Element Method for Engineers, Pentech Press Lon
don,
1978. Computational Mechanics Publications, Boston, 1984.
[7] Brebbia, C. A., Telles, J. ? F. and Wrobel, L. C. Boundary Element Technique
s, SpringerVerlag, Berlin and NY, 1984.
[8] Jaswon. M. A. and Symm, G. T. Integral Equation Methods in Potential Theory
and
Elastostatics, Academic Press, London, 1977.
[9] Dominguez, J. and Abascal, R. On Fundamental Solutions for the Boundary Inte
gral
liquations Method in Static and Dynamic Elasticity, Engineering Analysis, 1 128134,
1984.
[10] llartmann, ! ". 'filastostatics'. Progress in Boundary Element Methods, C. A.
Brebbia,
Ed., Pentech Press, London, 1981.
[II] Kermanidis, T. A. Numerical Solution for Axially Symmetrical Elasticity Pro
blems,
Journal of Solids and Structures, II, 493-500, 1975.
[ 12] Mayr, M. Ein Integralgleichungsverfahren Losung Rotationssymetrischer Elas
tizitatsprobleine. Dissertation, Technical University Munchen, 1975.
[13] Cruse, T. A., Snow, D. A. and Wilson, R. B. Numerical Solutions in Axisymme
tric
Elasticity, Computers and Structures, 8, 445-451, 1977.
[14] Gome/-Lent, M. S., Dominguez, J. and Alarcon, E. On the Use of a 3-D Fundam
ental
Solution for Axisymmetric Steady-State Dynamic Problems, Boundary Elements VII,
Proceedings of the 7ih International Conference, Italy, ? A. Brebbia and G. Maie
r
Eds, Springer-Verlag, Berin and NY, 1985.
[15] Wilson, E. Structural Analysis of Axisymmetric Solids, AIAA Journal, 3,2269
-2274,
1965.
[ 16] Brebbia, ? A. and Nardini, D. Dynamic Analysis in Solid Mechanics by an Al
ternative
Boundary Element Formulation, Int. Jnl. Soil Dynamics and Earthquake Engineering
,
2, 1983.
[ 17] Nardini, D. and Brebbia, C. A. Boundary Integral Formulation of Mass Matri
ces for
Dynamic Analysis, Chapter 7 in Topics in Boundary Element Research, Vol. 2 (Ed.
C. A.
Brebbia), Springer-Verlag, Berlin and NY, 1985.
[18] Wrobel, L. C, Telles, J. C. F. and Brebbia, C. A. A Dual Reciprocity Bounda
ry
Element Formulation for Axisymmetric Diffusion Problems, in BETECH/86 (Eds. J. J
.
Connor and ? A. Brebbia), Computational Mechanics Publications, Southampton
and Boston, 1986.
[19] Wrobel. I... Brehbia, C. A. and Nardini, D. Analysis of Transient Thermal P
roblems
in the HI-ASY System, in BKTKC ?/?6 (Eds. J. J. Connor and ? A. Brebbia),
Computational Mechanics Publications, Southampton and Boston, 1986.
[20] Brebbia, C. A. and Wrobel, L. Nonlinear Transient Thermal Analysis using th
e Dual
Reciprocity Method, in Boundary Element Techniques: Applications in Stress Analy
sis
(ii) The body and surface forces acting on the cylindrical surface have no x3
component and are independent of x3.
In this case in addition to ?3 = ?, one can assume that the in-plane displacemen
ts
Mt and u2 are independent of x3. Hence
"i =Mi(x,,x2); u2 = u2(x1,x2); ?? = 0 D.3)
This means that some of the resulting strains will also be zero, i.e.
e33 = e31 = 832 = 0
D.4)
and the others independent of x3, i.e. en(x1, x2), e22(x1, x2), e12(x!, x2).
For this case <x33 / ? and can be determined from the value of the other
components.
Constitutive Relations
One can now expand the three dimensional stress-strain relationships for an
isotropic body in terms of the non-zero components for plane strain or plane
stress, i.e.
1 ,
?
?ii=-(<7u-W22-vff33)
L
e22 = -(-VCTi1+ff22-V<T33)
(
+ )
212
Chapter 4 Two Dimensional Elastostatics
The relationships for plate stretching can be obtained by setting a33 = 0 in
D.5) and considering only the planar components en,?22 and e12, i.e.
?22 = -(-VOU+ff22)
D-6)
The value of e33 can be obtained a posteriori from the third relationship in D.5
), i.e.
?33 = -(-Vffii-w22)
D-7)
Equations D.6) can be inverted to produce,
?"!! =~ 2(?H+V?22)
1-V2
?
D-8)
^22 = 7 i (ven+e22)
l-vz
The plane strain equations can be found by first eliminating e33 from D.5), i.e.
e33 = -(-vffu-V(T22 + <733)s0 D.9)
which gives
<*?? = v(ffi + o22)
D.10)
Substituting for c733 in equation D.5) one obtains the following expressions,
eu=-[(l-v2)ffn-v(H-v)ff22]
?22 = k-v(l+v)<T11 + (l-v2),722]
D.11)
t
[d-v)eu
0-v)e22]
D.12)
4.3. Boundary Element Formulation
213
The inverse of these expressions is
E
ffu~(l + v)(l-2v)
?
22~(\+v)(\-2v)
o-12 = 2Ge12
The value of e33 if required can be obtained from D.10).
Notice that it is sometimes more convenient to express the first two relationshi
ps
above using Lame's constant X = v?/[(l + v)(l
2v)], i.e.
?
v
" "
D.13)
?
v
One can pass from equations D.8) for plate stretching to equations D.12) for
plane strain simply by replacing ? and v in the first equation by two equivalent
values, ?' and v', such that
F.
v
v' = - D.14)
?
v
1 - v2' 1
v
where the value of G remains the same.
This interesting relationship means that one can implement a plate stretching
program and by transforming the elastic constant data in accordance to D.14) sol
ve
also a plane strain problem. This is done in finite element analysis. Conversely
in
boundary elements one works with plane strain problems as the fundamental
solution is known for this type of problem and then plate stretching problems ca
n
be solved using the inverse relationship to D.14), i.e.
' = (l-v2)?; v' =
D.15)
1 + v
4.3 Boundary Element Formulation
The basic relationships for boundary elements in elastostatics have been develop
ed
in Chapter 3. In what follows some of the corresponding equations will be review
ed
and specialized for two dimensional cases with a view to applying them in two
computer codes, one using constant element and the other quadratic elements,
the latter more appropriate for elasticity problems. The development of an
214
Chapter 4 Two Dimensional Elastostatics
elastostatics linear code if required, is left to the reader as it follows simil
ar lines as
the linear code for potential problems.
One can start with the integral expression C.79), i.e.
4A +1 ??? dT = | pkutk dT + I bkufk dQ D.16)
?
?
?
where the fundamental solution has been assumed to satisfy the following equatio
n,
G^. + ?' = 0
D.17)
Equation D.16) applies for points on the boundary or internal points (with
c\k = 8lk). Smooth surfaces give c\k = %dlk and corners produce a different type
of
c\k array as will be discussed shortly.
The fundamental solution for an isotropic material in plane strain has been
given in Chapter 3 but will be repeated here for completeness.
8ttGA v)|
= _.___ _ \Z.\(i-2v)8kl + 2-^
>
D.18)
47?A v)r \_dn I
dxk dxt)
pfk and ufk represent the tractions and displacements in the ? direction due to
a
unit load in the / direction acting at 'f.
One can now write equation D.16) in matrix form by defining the following
arrays. The fundamental solution components can be written as two 2x2 matrices
with elements ?& and pfk, i.e.
U* = K <? *Jrtx phi
D19)
L"*! 2J LP*1 P*2J
The displacement, tractions and body forces vectors are
Hence the basic equation becomes,
cV + J p*u dT = I u*p dr + J u*b dil
D.21)
?
?
?
where the u' defines the displacements at the - internal or boundary - point 1
4.4. Constant Element Formulation
215
where the load is applied, c' is a 2 x 2 array of constant which values depend o
n
the type of point under consideration. If 'f is an internal point
i ?1
c=Lo i
D22)
If T is a boundary point on a smooth surface then,
-g a
a
If 'f is a corner, we will have
D.24)
where the c,* values will depend on the type of corner under consideration as we
will see shortly.
4.4 Constant Element Formulation
Consider now that the surface of the boundary under study is discretized using
constant elements (figure 4.2). This implies that the values of u and p are assu
med
to be constant on each element and equal to the value at the mid-node of the
element. One can also discretize the interiors of the domain in a number of cell
s
which are required for the integration of the body force term in D.21). These ce
lls
are used only for numerical integration of the body force terms and in certain
cases - as seen in Chapter 3 - they can be taken to the boundary. Consider here
that the body fi is discretized into N boundary elements and M internal cells,
hence formula D.21) can be written as,
<*' + ? Ij p* dr\vi = ? 11u*dr}pj' +1 \i *bd
J=l (Xj J
j=l (.?,
J
s=l (p.
The above equation corresponds to the particular node 'i' where the unit forces
are assumed to be acting.
Notice that terms such as J u dT and J p dT relate node '? to the element or
r,
r, .
node'/. They produce a type of influence coefficient. After integration the inte
grals
produce two 2x2 submatrices called ft'7 and GiJ. Numerical integration of the
body forces term can be carried out as follows,
I jf u*bdn}= t j I (u*b)pwpJQs= I B"
D.26)
s=l (.?,
J s=l lp=l
)
s=l
216
Chapter 4 Two Dimensional Elastostatics
Nodes
Elements
- Pi
Figure 4.2 Two-dimensional body divided into boundary elements and internal
cells
Notice that this produces two components of BIS, i.e. B? and ? after the numeric
al
integration has taken place. wp are the weighting coefficients and ils the area
of
the cell under consideration. The function (u*b) has to be calculated at the p
integration points, where p varies from 1 to r.
Equation D.25) can now be written
?
N
?
?'?'? X A'V= Z G'V+ ? Bis
j=l
7=1
s=l
D.27)
This equation relates the values of u at node 'F with the values of u's and p's
at
all the nodes on the boundary, including V. Notice that in this case - smooth
CLOSE (INP)
CLOSE (IPR)
STOP
END
2. Routine INPUTEC
This subroutine reads the input required by the code. A file whose name is reque
sted
from the user by the main program will contain the following lines of free forma
t
input.
(i) Title Line Contains the title of the problem.
(ii) Basic parameter line Contains the number of elements, number of
surfaces, last nodes of each different surface, the shear modulus and Poisson
ratio or fictitious Poisson ratio for plane stress problems v/(l + v).
(iii) Extreme Points of Boundary Elements Lines The coordinates of the
extreme point of an element are read in the counterclockwise direction for
external surfaces and clockwise for internal ones. The coordinates are read
in free FORMAT.
(iv) Boundary Conditions Lines As many lines as boundary nodes giving the
values of the known variable in xt and x2 directions. The variables are
displacements if KODE = 0 or tractions if KODE = 1.
(v) Internal Points Coordinate Lines Internal nodes are defined, each with the
xtx2 coordinates of the point. The coordinates are read in free FORMAT.
The subroutine prints first the name of the job and the basic parameters. Then
the coordinates of the extreme points of the elements and the boundary condition
s
given by node number, codes and prescribed values are printed. The internal poin
t
coordinates will be printed in the subroutine OUTPTEC. Its listing is as follows
:
SUBROUTINE INPUTEC(CX,CY,X,Y,KODE,FI)
?
? PROGRAM 29
?
4.5. Elastostatics Code using Constant Elements (ELCONBE)
223
CHARACTEK*80 TITLE
DIMENSION CXA),CYA),XA),YA),KODEA),FIA)
COMMON N,L,NCE),M,GE,XNU,INP,IPR
?
? ?= NUMBER OF BOiraDARY NODES (= NUMBER OF ELEMENTS)
? L= NUMBER OF INTERNAL POINTS WHERE DISPLACEMENT AND STRESS
? ARE CALCULATED
? ?= NUMBER OF DIFFERENT BOUNDARIES
? NC(I)= LAST NODE OF BOUNDARY I
? GE= SHEAR MODULUS
? XNU= POISSON MODULUS
?
WRITE(IPR,100)
100 FORMAT(/' ',79('*'))
?
? READ JOB TITLE
?
READ(INP,'(A)') TITLE
WRITE(IPR,'(A)') TITLE
?
? READ NUMBER OF NODES, INTERNAL POINTS AND DIFFERENT BOUNDARIES;
? READ LAST NODES OF THESE BOUNDARIES AND MATERIAL PROPERTIES
?
READ(INP,*)N,L,M, <NC(K) , K-l, 5) ,GE,XNU
WRITE(IPR,300)N,L,GE,XNU
? PROGRAM 30
?
? THIS SUBROUTINE COMPUTES THE G AND H MATRICES AND
? FORMS THE SYSTEM OF EQUATIONS A X = F
?
DIMENSION G(NX,NX),H(NX,NX)
DIMENSION X(l),YA),XMA),YMA),FIA)
DIMENSION KODE(l),DFIA)
COMMON N,L,NCE),M,GE,XNU,INP,IPR
?
? COMPUTE THE NODAL COORDINATES AND STORE IN ARRAYS XM AND YM
?
X(N+1)=XA)
Y(N+I)=YA)
DO 10 1=1,N
()))/
10 YM(I)=(Y(I)+Y(I+l))/2
IF(M1)
12
XM(NC(l))=(X{NC(l))+X(l))/2
YM(NC(l))=(Y(NC(l))+Y(l))/2
DO 13 K=2,M
XM(NC(K))=(X(NC(K))+X(NC(K-l)+l))/2
13
YM(NC(K))=(Y(NC(K))+Y(NC(K-l)+l))/2
?
? COMPUTE THE COEFICIENTS OF G AND H MATRICES
?
15 DO 30 1=1,N
DO 30 J=1,N
17
IF(J-NCA)I9,18,19
18
KK=1
GO TO 23
19
DO 22 K=2,M
IF(J-NC(K)J2,21,22
21
KK=NC(K-1)+1
GO TO 23
22
CONTINUE
16 KK=J+1
23
IF(I-JJ0,2S,20
4.5. Elastostatics Code using Constant Elements (ELCONBE)
225
20 CALL EXTINEC(XM(I),YM(I) ,X(J),Y(J) , X(KK) , Y(KK) , H( B*1-1) , B*J-1) ) ,
1H(B*1-1),B*J)),H(B*1),B*J-1)),H(B*1),B*J)),G(B*1-1),
2B*J-1)),G(B*1-1),B*J)),G(B*1),<2*J)))
G(B*1),B*J-1))=G(B*1-1),B*J))
GO TO 26
25
CALL LOCINEC(X(J),Y(J),X(KK),Y(KK),G(B*1-1),B*J-1)),
1G(B*1-1),B*J)),G(B*1),B*J)))
H(B*1-1),B*J-l))=0.5
H(B*I),B*J))=0.5
H(B*I-l),B*J))=0.
H(B*I),B*J-l))=0.
G( B*1) , B*J-1) )=G( B*1-1) , B*J) )
26
CONTINUE
30 CONTINUE
?
? REORDER THE COLUMNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE
? WITH THE BOUNDARY CONDITIONS AND FORM SYSTEM MATRIX A WHICH
? IS STORED IN G
?
NN=2*N
DO 50 J=1,NN
IF(KODE(J)L3,43,40
4 0 DO 42 1=1,NN
CH=G(I,J)
G(I,J)=-H(I,J)
42
H(I,J)=-CH
GO TO 50
43
DO 45 1=1,NN
45 G(I,J)=G(I,J)*GE
50 CONTINUE
?
? FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN DFI
?
DO 60 1=1,NN
DFI(I)=0.
DO 60 J=1,NN
DFI(I)=DFI(I)+H(I,J)*FI(J)
60 CONTINUE
RETURN
END
4. Routine EXTINEC
This routine computes the G'J and H'J submatrices (for 1?]) using numerical
integration. Notice that G'J is symmetric but H'J is not. The values of the coef
ficients
are computed using four points Gaussian quadrature.
?
SUBROUTINE EXTINEC(XP,YP,X1,Y1,X2,Y2,H11,H12,H21,H22,G11,G12,G22)
?
? PROGRAM 31
?
? THIS SOUBROUTINE COMPUTES THE G AND H MATRICES
? COEFFICIENTS THAT RELATE A COLLOCATION POINT WITH A DIFFERENT
? ELEMENT USING GAUSS QUADRATURE
?
? DIST= DISTANCE FROM THE COLOCATION POINT TO THE
? LINE TANGENT TO THE ELEMENT
? RA= DISTANCE FROM THE COLOCATION POINT TO THE
? GAUSS INTEGRATION POINT AT THE BOUNDARY ELEMENT
?
DIMENSION XCOD),YCOD),GID),OMED) '
COMMON N,L,NCE) ,M,GE,XNU, INP, IPR
DATA GI/0.86113631,-0.86113631,0.33998104,-0.33998104/
DATA OME/0.34785485,0.34785485,0.65214515,0.65214515/
?
AX=(X2-Xl)/2
BX=(X2+Xl)/2
226
Chapter 4 Two Dimensional Elastostatics
AY=(Y2-Yl)/2
BY=(Y2+Yl)/2
ETA1=(Y2-Y1)/B*SQRT(AX**2+AY**2))
ETA2=(X1-X2)/B*SQRT(AX**2+AY**2))
?
? COMPUTE THE DISTANCE FROM THE POINT TO THE LINE OF THE ELEMENT
?
IF(AXI0,20,10
10 TA=AY/AX
DIST=ABS( (TA*XP-YP+Y1-TA*X1)/SQRT(TA**2+1))
GO TO 3 0
20 DIST=ABS(XP-X1)
?
? DETERMINE THE DIRECTION OF THE OUTWARD NORMAL
?
30
SIG=(X1-XP)*(Y2-YP)-(X2-XP)*(Yl-YP)
IF(SIGK1,32,32
31
DIST=-DIST
32
Hll=0.
H12=0.
H21=0.
H22=0.
Gll=0.
G12=0.
G22=0.
?
? COMPUTE G AND H COEFFICIENTS
?
DE=4*3.141592*A-XNU)
DO 40 1=1,4
XCO(I)=AX*GI(I)+BX
YCO(I)=AY*GI(I)+BY
RA=SQRT( (XP-XCO(I)) **2+(YP-YCO(I) ) **2)
RD1=(XCO(I) -XP)/RA
RD2=(YCO(I)-YP)/RA
Gll=Gll+(C-4*XNU)*ALOGA./RA)+RD1**2)*OME(I)*SQRT(AX**2+AY**2)/B
1*DE*GE)
G12=G12+RD1*RD2 *OME(I)*SQRT(AX**2+AY**2)/B *DE*GE)
G22=G22+(C-4*XNU)*ALOG(l./RA)+RD2**2)*OME(I)*SQRT(AX**2+AY**2)/B
1*DE*GE)
H11=H11-DIST*(A-2*XNU)+2*RD1**2)/(RA**2*DE)*OME(I)*SQRT(AX**2+AY*
1*2)
H12=H12-(DIST*2*RD1*RD2/RA+A-2*XNU)*(ETA1*RD2-ETA2*RD1))*OME(I)*S
1QRT(AX**2+AY**2)/(RA*DE)
H21=H21-(DIST*2*RD1*RD2/RA+A-2*XNU) * (ETA2*RD1-ETA1*RD2)) *OME (I) *S
1QRT(AX**2+AY**2)/(RA*DE)
40 H22=H22-DIST*(A-2*XNU)+2*RD2**2)*OME(I)*SQRT(AX**2+AY**2)/(RA**2*
IDE)
RETURN
END
5. Routine LOCINEC
It calculates the G" submatrix in accordance with formulae D.42) to D.44), which
give the analytical results.
_______
Q
_
___ ...._-... -.________
SUBROUTINE LOCINEC (XI, Yl, X2, Y2 ,Gil,G12, G22)
?
? PROGRAM 32
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE MATRIX G COEFFICIENTS
? THAT RELATE AN ELEMENT WITH ITSELF
?
COMMON N,L,NCE),M,GE,XNU,INP,IPR
?
AX=(X2-Xl)/2
AY=(Y2-Yl)/2
SR=SQRT(AX**2+AY**2)
4.5. Elastostatics Code using Constant Elements (ELCONBE)
227
DE=4*3.141592*GE*A-XNU)
G11=SR*(C-4*XNU)*A-ALOG(SR))+(X2-X1)**2/D*SR**2))/DE
G22=SR*(C-4*XNU)*A-ALOG(SR))+(Y2-Y1)**2/D*SR**2))/DE
G12=(X2-X1)*(Y2-Y1)/D*SR*DE)
RETURN
END
6.
Routine SLNPD
This is the same solver as used in previous codes.
7.
Routine INTEREC
This subroutine computes the displacements and stresses at internal points. Firs
t
it reorders the vectors DFI and FI. Notice that the first contains the values of
X
in equation D.31) as given by the solver and the second the boundary conditions.
After reorganization the boundary displacements are all stored in FI and the
tractions in DFI.
Formulae D.33) and D.35) are applied to compute internal stresses and
displacements. This requires the integration of terms with S and D coefficients
which are carried out numerically in a routine called SIGMAEC.
The FORTRAN Listing is as follows:
?
SUBROUTINE INTEREC(FI,DFI,KODE,CX,CY,X,Y,SSOL,DSOL)
?
? PROGRAM 33
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE STRESS AND DISPLACEMENT
? COMPONENTS AT INTERNAL POINTS
?
DIMENSION CXA),CY(I),SSOL(I),DSOL(l)
DIMENSION FIA),DFIA),KODEA),XA),YA)
COMMON N,L,NCE),M,GE,XNU,INP,IPR
?
? REARRANGE FI AND DFI ARRAYS TO STORE ALL THE VALUES
? OF THE DISPLACEMENT IN FI AND ALL THE VALUES OF THE TRACTIONS IN DFI
?
NN=2*N
DO 20 I=1,NN
IF(KODE(I)) 15,15,10
10 CH=FI(I)
DFI(I)-=CH
GO TO 20
15 DFI(I)=DFI(I)*GE
20 CONTINUE
?
? COMPUTE THE VALUES OF STRESSES AND DISPLACEMENTS
? AT INTERNAL POINTS.
?
IF(L.EQ.O) GO TO 50
DO 40 K=1,L
DSOLB*K-1)=0.
DSOLB*K)=0.
SSOLC*K-2)=0.
SSOLC*K-1)=0.
SSOLC*K)=0.
DO 30 J=1,N
IF(M-1J8,28,22
22 IF(J-NCA)J4,23,24
228
Chapter 4 Two Dimensional Elastostatics
23
KK=1
GO TO 29
24
DO 26 LK=2,M
IF(J-NC(LK)J6,25,26
25
KK=NC(LK-1)+1
GO TO 29
26
CONTINUE
28
KK=J+1
29
CALL EXTINEC(CX(K),CY(K),X(J),Y(J),X(KK),Y(KK),H11,H12,H21,H22,
1 G11,G12,G22)
DSOLB*K-1)=DSOLB*K-1)+DFIB*J-1)*G11+DFIB*J)*G12-FIB*J-1)*H111FIB*J)*H12
DSOLB*K)=DSOLB*K)+DFIB*J-1)*G12+DFIB*J)*G22-FIB*J-1)*H21-FIB
1*J)*H22
CALL SIGMAEC(CX(K),CY(K),X(J),Y(J),X{KK),Y(KK)/D111,D211,D112,
1 D212,D122,D222,S111,S211,S112,S212,S122,S222)
SSOLC*K-2)=SSOLC*K-2)+DFIB*J-l)*Dlll+DFIB*J)*D211-FIB*J-l)*Sl
111-FIB*J)*S211
SSOLC*K-1)=SSOLC*K-1)+DFIB*J-1)*D112+DFIB*J)*D212-FIB*J-1)*S1
112-FIB*J)*S212
30
SSOLC*K)=SSOLC*K)+DFIB*J-1)D122+DFIB*J)*D222-FIB*J-1)*S122-F
1IB*J)*S222
40 CONTINUE
50 RETURN
END
8. Routine SIGMAEC
This subroutine computes the integrals with S and D terms described in formulae
D.35) and needed to calculate the internal stresses. The integrations are carrie
d
out using a four points Gaussian integration scheme.
._.._ ____ ..___ _._____ .-.__
..._.___ ~~
^ _ _. ___^_ _
SUBROUTINE SIGMAEC(XP,YP,XI,Yl,X2,Y2,Dill,D211,D112,D212,D122,
1 D222,S111,S211,S112,S212,S122,S222)
?
? PROGRAM 34
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE S AND D MATRICES
? USING GAUSS QUADRATURE IN ORDER TO COMPUTE THE STRESSES
? AT ANY INTERNAL POINT
?
? RA= DISTANCE FROM THE POINT TO THE GAUSS INTEGRATION POINTS
? ON THE BOUNDARY ELEMENTS
? DIST= DISTANCE FROM THE POINT TO THE LINE TANGENT
? ?? THE ELEMENT
? RD1,RD2= DERIVATIVES OF RA
? ???1,???2= COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
?
DIMENSION XCOD),YCOD),GID),OMED)
COMMON N,L,HCE),M,GE,XNU,INP,IPR
DATA GI/O.86113631,-0.86113631,0.33998104,-0.33998104/
DATA OME/0.34785485,0.34785485,0.65214515,0.65214515/
?
AX=(X2-Xl)/2
BX=(X2+Xl)/2
AY=(Y2-Yl)/2
BY=(Y2+Yl)/2
ETA1=(Y2-Y1)/B*SQRT(AX**2+AY**2))
ETA2=(X1-X2)/B*SQRT(AX**2+AY**2))
?
? COMPUTE THE DISTANCE FROM THE POINT TO THE LINE OF THE ELEMENT
?
IF(AXI0,20,10
10 TA=AY/AX
DIST=ABS((TA*XP-YP+Y1-TA*X1)/SQRT(TA**2+1))
GO TO 30
20 DIST=ABS(XP-X1)
4.5. Elastostatics Code using Constant Elements (ELCONBE)
229
__
?
? DETERMINE THE DIRECTION OF THE OUTWARD NORMAL
?
30
SIG=(X1-XP)*(Y2-YP)-(X2-XP)*(Vl-YP)
IF(SIGK1,32,32
31
DIST=-DIST
32
Dlll=0.
D211=0.
D112=0.
D212=O.
D122=0.
D222=0.
Slll=0.
S211=0.
S112=0.
S212=0.
S122=0.
S222=0.
?
? COMPUTE D AND S COEFFICIENTS
?
FA=1-4*XNU
AL=1-2*XNU
DE=4*3.141592*A-XNU)
DO 40 1=1,4
XCO(I)=AX*GI(I)+BX
YCO(I)=AY*GI(I)+BY
RA=SQRT((XP-XCO(I))**2+(YP-YCO(I))**2)
RD1=(XCO(I)-XP)/RA
RD2=(YCO(I)-YP)/RA
D111=D111+(AL*RD1+2*RD1**3)*OME(I)*SQRT(AX**2+AY**2)/(DE*RA)
D211=D211+B*RD1**2*RD2-AL*RD2)*OME(I)*SQRT(AX**2+AY**2)/(DE*RA)
D112=D112+(AL*RD2+2*RD1**2*RD2)/(DE*RA)*OME(I)*SQRT(AX**2+AY**2)
D212=D212+(AL*RD1+2*RD1*RD2**2)/(DE*RA)*OME(I)*SQRT(AX**2+AY**2)
D122=D122+B*RD1*RD2**2-AL*RD1)/(DE*RA)*OME(I)*SQRT(AX**2+AY**2)
D222=D222+(AL*RD2+2*RD2**3)/(DE*RA)*OME(I)*SQRT(AX**2+AY**2)
S111=S111+B*DIST/RA*(AL*RD1+XNU*2*RD1-4*RD1**3)+4*XNU*ETA1*RD1**2
1+AL* B*ETA1*RD1**2+2*ETA1)-FA*ETA1) *2*GE/(DE*RA**2) *0ME(I) *SQRT(AX
2**2+AY**2)
S211=S211+B*DIST/RA*(AL*RD2-4*RD1**2*RD2)+4*XNU*ETA1*RD1*RD2+AL*2
1*ETA2*RD1**2-FA*ETA2)*2*GE/(DE*RA**2)*OME(I)*SQRT(AX**2+AY**2)
S112=S112+B*DIST/RA*(XNU*RD2-4*RD1**2*RD2)+2*XNU*(ETA1*RD2*RD1+ET
1A2*RD1**2)+AL*B*ETA1*RD1*RD2+ETA2))*2*GE/(DE*RA**2)*OME(I)*SQRT(A
2X**2+AY**2)
S212=S212+B*DIST/RA*(XNU*RD1-4*RD1*RD2**2)+2*XNU*(ETA1*RD2**2+ETA
12*RD1*RD2)+AL* B*ETA2*RD1*RD2+ETA1)) *2*GE/(DE*RA**2) *OME (I) *SQRT(A
2X**2+AY**2)
S122=S122+B*DIST/RA*(AL*RD1-4*RD1*RD2**2)+4*XNU*ETA2*RD1*RD2+AL*2
1*ETA1*RD2**2-FA*ETA1)*2*GE/(DE*RA**2)*OME(I)*SQRT(AX**2+AY**2)
40 S222=S222+B*DIST/RA*(AL*RD2+2*XNU*RD2-4*RD2**3)+4*XNU*ETA2*RD2**2
1+AL*B*ETA2*RD2**2+2*ETA2)-FA*ETA2)*2*GE/ (DE*RA**2) *0ME(I) *SQRT(AX
2**2+AY**2)
RETURN
END
9. Routine OUTPTEC
This subroutine prints the results in the following order.
(ii) Boundary nodes values with coordinates xtx2, values of u,m2 displacements
and pip2 tractions,
(ii) Internal node results with coordinates xxx2, values of ulu2 displacements
and crllal2a22 stresses.
Notice that stresses on the boundary are not produced. If required they will nee
d
to be computed using formulae C.5) with the derivatives of displacements
230
Chapter 4 Two Dimensional Elastostatics
tangential to the surface computed by considering the nodes of displacements and
adjacent elements and applying a finite different type approach.
The listing of OUTPT is as follows:
(^__
_ __ -_~.__w
____
__ _....
____ __ _ .__.._...
_ __
SUBROUTINE OUTPTEC(XM,YM,FI,DFI,CX,CY,SSOL,DSOL)
?
? PROGRAM 35
?
? THIS SUBROUTINE PRINTS THE VALUES OF THE DISPLACEMENTS
? AND TRACTIONS AT BOUNDARY NODES. IT ALSO PRINTS THE VALUES
? OF DISPLACEMENTS AND STRESSES AT INTERNAL POINTS
?
DIMENSION XMA),YMA),FIA),DFIA)
DIMENSION CXA),CYA),SSOLA),DSOL(l)
COMMON N,L,NCE) ,M,GE, XNU, INP, IPR
?
WRITE(IPR.IOO)
100 FORMATC ',79('*')//lX,'RESULTS'//2X,'BOUNDARY NODES'//6X
1,'X',12X,'Y',9X,'DISPL. X',5X,'DISPL. Y',4X,
2'TRACTION X',3X,'TRACTION Y'/j
DO 10 1=1,N
10 WRITE(IPR,200) XM(I),YM(I),FIB*I-1),FIB*1),DFIB*1-1),DFIB*1)
200 FORMATFAX,E12.5))
?
IF(L.EQ.0.) GO TO 30
WRITE(IPR,300)
300 F0RMAT(//2X,'INTERNAL POINTS DISPLACEMENTS'//8X,'X',15X,'Y',10X,
1'DISPLACEMENT X',5X,'DISPLACEMENT Y')
DO 20 K=1,L
20 WRITE(IPR,400)CX(K) ,CY(K) ,DSOLB*K-1) ,DSOLB*K)
WRITE(IPR,350)
350 FORMAT(//2X,'INTERNAL POINTS STRESSES'//8X,'X',15X,'Y',12X,
l'SIGMA X',10X,'TAU XY',9X,'SIGMA Y')
DO 25 K=1,L
25 WRITE(IPR,450) CX(K),CY(K),SSOLC*K-2),SSOLC*K-1),SS0LC*K)
400 FORMATBBX,E14.7),2EX.E14.7))
450 FORMATEBX,E14.7))
30 WRITE(IPR,500)
500 FORMATC ',79 ('*'))
RETURN
END
Example 4.1
Figure 4.4 describes a circular cavity under internal pressure in an infinite me
dium.
Numerical results will be compared against the known analytical solution, first
using constant elements and then with the quadratic elements of section 4.7 and
4.8.
The boundary is divided here into 24 constant elements and 5 internal nodes
are defined at which displacements and stresses will be found.
Notice that in order to stop all rigid body movements one can suppress the
displacements in the x2 directions at node 18.
The input needed to run the ELCONBE code is as follows.
CIRCULAR CAVITY (DATA)
CIRCULAR CAVITY UNDER INTERNAL PRESSURE B4 CONSTANT ELEMENTS)
24 5 000000 94500. 0.1
-.3916 -2.9743
-1.1481 -2.7716
-1.8263 -2.3801
-2.3801 -1.8263
-2.7716 -1.1480
-2.9743 -.3916
-2.9743 .3916
4.5. Elastostatics Code using Constant Elements (ELCONBE)
231
-2.7716
-2.3801
-1.8263
-1.1480
.
1
1
2
2
2
2
2
2
1
1
.
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
0
4
.3916
3916 2.
.1481 2.
.8263 2.
.3801 1.
.7716 1.
1.1481
1.8263
2.3801
2.7716
2.9743
9743
7716
3801
8263
1480
.9743 .3916
.9743 -.
.7716 -1
.3801 -1
.8263 -2
.1480 -2
3916 -2
-25.88
-50.
-70.71
-86.6
-96.59
-100.
-96.59
-86.6
-70.71
-50.
-25.88
0.
25.88 1
50. 1
70.71 1
86.6 1
96.59 1
100. 0
96.59 1
86.6 1
70.71 1
50. 1
25.88 1
0. 1
3916
.1481
.8263
.3801
.7716
.9743
1 -96.59
1 -86.6
1 -70.71
1 -50.
1 -25.88
1 0.
1 25.88
1 50.
1 70.71
1 86.6
1 96.59
1 100.
96.59
86.6
70.71
50.
25.88
0.
-25.88
-50.
-70.71
-86.6
-96.59
-100.
. 0. 2.82843 2.82843 -4.
0. 6. 0. 10. 0.
This produces the output given below. Notice that the results are very close to
the theoretical results as shown by the values of radial stresses at internal po
ints.
Stresses and displacements decay with increasing distance from the cavity as is
to
be expected. Radial and hoop stresses at internal points have the same absolute
value but with different signs as expected.
Table 4.1 Radial Stresses at Internal Points. 24 Constant Elements
Discretization
Distance to the Constant boundary
Elasticity
centre of the cavity element discretization
theory
4
6
10
20
50
200
1000
-57.234
-25.295
-9.106
-2.276
-0.364
-0.227 x 10"'
-0.991 x 10~3
-56.250
-25.000
-9.000
-2.250
-0.360
-0.225 x 10"'
-0.9 x 10
232
Chapter 4 Two Dimensional Elastostatics
The computer output is as follows,
CIRCULAR CAVITY (OUTPUT)
*****************************************************************************
CIRCULAR CAVITY UNDER INTERNAL PRESSURE B4 CONSTANT ELEMENTS)
DATA
NUMBER OF BOUNDARY ELEMENTS = 24
NUMBER OF INTERNAL POINTS = 5
SHEAR MODULUS = 0.9450000E+05
POISSON RATIO = 0.1000000E+00
COORDINATES OF THE EXTREME POINTS OF' THE BOUNDARY ELEMENTS
POINT
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
-0.3916000E+00
-0.1148100E+01
-0.1826300E+01
-0.2380100E+01
-0.2771600E+01
-0.2974300E+01
-0.2974300E+01
-0.2771600E+01
-0.23801O0E+O1
-0.1826300E+01
-O.1148000E+O1
-0.3916000E+00
0.3916000E+00
0.1148100E+01
0.1826300E+01
0.2380100E+01
0.2771600E+01
0.2974300E+01
0.2974300E+01
0.2771600E+01
0.2380100E+01
O.1826300E+01
0.1148000E+01
0.3916000E+00
-0.2974300E+01
-0.2771600E+01
-0.2380100E+01
-O.18263OOE+01
-0.1148000E+01
-0.3916000E+00
0.3916000E+00
0.1148100E+01
0.1826300E+01
0.2380100E+01
0.2771600E+01
0.2974300E+01
O.29743O0E+01
0.2771600E+01
0.2380100E+01
0.1826300E+01
0.1148000E+01
0.3916000E+00
-0.3916000E+00
-0.1148100E+01
-0.1826300E+01
-O.238O1O0E+01
-0.2771600E+01
-0.2974300E+01
BOUNDARY CONDITIONS
PRESCRIBED VALUE
NODE X DIRECTION
1
-0.2588000E+02
2
-0.5000000E+02
3
-0.7071000E+02
4
-0.8660000E+02
5
-0.9659000E+02
6
-0.1000000E+03
7
-0.9659000E+02
8
-0.8660000E+02
9
-0.7071000E+02
10
-0.5000000E+02
11
-0.2588000E+02
12
0.OOO00OOE+O0
13
0.2588000E+02
14
0.5000000E+02
15
0.7071000E+02
16
0.8660000E+02
17
0.9659000E+02
18
0.1000000E+03
19
0.9659000E+02
20
0.8660000E+02
21
0.7071000E+02
22
0.5000000E+02
23
0.2588000E+02
24
0.0000000E+00
****************************************************************************<
CODE
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
0
PRESCRIBED VALUE
Y DIRECTION
-0.9659000E+02
-0.8660000E+02
-0.7071000E+02
-0.5000000E+02
-0.2588000E+02
0.0000000E+00
0.2588000E+02
0.5000000E+02
0.7071000E+02
0.8660000E+02
0.9659000E+02
0.1000000E+03
0.9659000E+02
0.8660000E+02
0.7071000E+02
0.5000000E+02
0.2588000E+02
0.0000000E+00
-0.2588000E+02
-0.5000000E+02
-0.7071000E+02
-0.8660000E+02
-0.9659000E+02
-0.1000000E+03
CODE
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
4.6. Linear Elements
233
RESULTS
BOUNDARY NODES
-0
0.76985E+00
0.14872E+01
0.21032E+01
0.25759E+01
0.28730E+01
0.29743E+01
0.28730E+01
25759E+01
0.21032E+01
0.14872E+01
0.76980E+00
0.00000E+00
76985E+00
14872E+01
21032E+01
25759E+01
28730E+01
29743E+01
28730E+01
25759E+01
21032E+01
14872E+01
O.7698OE+OO
0.0000OE+0O
-0.28730E+01
0.25759E+01
-0.21032E+0X
0.14872E+01
0.76980E+00
0.00000E+00
0.76985E+00
0.14872E+01
O.21032E+01
0.25759E+01
0.28730E+O1
0.29743E+01
0.28730E+01
0.25759E+01
0.21032E+01
O.14872E+O1
0.76980E+00
O.00000E+0O
-0.76985E+00
-0.14872E+01
-0.21032E+01
257593+01
0.28730E+O1
0.29743E+01
-0
DISPL. X
-0.42449E-03
-0.82002E-03
.11597E-02
14204E-02
15842E-02
16401E-02
15842E-02
14204E-02
-0.H597E-02
-0.82001E-03
-0.42449E-03
0.O000OE+00
0.42449E-03
0.82002E-03
0.11597E-02
0.14204E-02
0.15842E-02
0.16401E-02
0.15842E-02
0.14204E-02
0.H597E-02
0.82001E-03
0.42449E-O3
0.00000E+00
DISPL. ?
-0.15842E-02
-0.14204E-02
-0.11597E-02
-0.82001E-O3
.42449E-03
.73851E-08
.42448E-03
.82002E-03
.11597E-02
0.14204E-02
0.15842E-02
0.16401E-02
0.15842E-02
0.14204E-02
0.11597E-02
0.82001E-03
0.42449E-03
0.000O0E+0O
-O.42448E-03
-0.82002E-03
-0.11597E-02
-0.14204E-02
-0.15842E-02
-0.16401E-02
TRACTION X TRACTION Y
-0.25880E+02
-0.50000E+02
-0.70710E+O2
-0.86600E+02
-0.96590E+02
-0.10000E+03
-0.96590E+02
-0.86600E+02
-0.70710E+02
-0.50000E+02
-0.25880E+02
0.14446E-02
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
0.96590E+02
0.86600E+02
0.70710E+02
0.50000E+02
0.2588OE+O2
-0.12447E-02
-0.96590E+02
-0.86600E+02
-0.70710E+02
-0.50000E+02
-0.25880E+02
O.OOOOOE+00
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
O.10000E+03
0.96590E+02
0.86600E+02
O.70710E+02
0.50000E+02
0.25880E+02
-0.20253E-02
-0.25880E+02
-0.50000E+02
-0.70710E+02
-O.866O0E+02
-O.96590E+02
-0.10000E+03
INTERNAL POINTS DISPLACEMENTS
0.4000000E+01
0.2828430E+01
-0.4000000E+01
0.6000000E+0X
0.1000000E+02
0.0000000E+00
0.2828430E+01
O.OOOOOOOE+00
0.0000000E+00
O.OOOOOOOE+00
DISPLACEMENT X
0.1204821E-02
0.8519205E-03
-0.1204821E-02
0.8029994E-03
0.4817977E-03
DISPLACEMENT Y
0.7319613E-08
0.8519267E-03
-0.4833055E-08
0.6300070E-08
0.6126356E-08
INTERNAL POINTS STRESSES
0.4000000E+01
0.2828430E+01
-0.4000000E+01
0.600000OE+01
0.1000000E+02
O.OOOOOOOE+00
0.2828430E+01
O.OOOOOOOE+00
O.OOOOOOOE+00
O.OOOOOOOE+00
SIGMA X
-0.5723444E+02
-0.5605986E-01
-0.5723446E+02
-0.2529478E+02
-0.9106032E+01
TAU XY
-0.5717874E-03
-0.5717675E+02
-0.6064773E-03
-0.7735193E-04
-0.1208484E-04
SIGMA Y
0.5711818E+
-0.5633205E0.57X1818E+
0.2529438E+
0.9105983E+
*****************************************************************************
4.6 Linear Elements
In this section we will consider the development of linear elements similar to t
hose
discussed in Chapter 2 section 5, for potential problems. The difference is that
now we need to interpolate two values for u's and two for p's (figure 4.5).
11 =
Pi 0 ?2 ?
? ?1 0 ?2
D.45)
234
Chapter 4 Two Dimensional Elastostatics
Internal points
Internal points
Figure 4.4 Circular cavity under internal pressure boundary element mesh and
internal point description
Node 2
Node 1
Figure 4.5 Linear element
W~L0 ?? ? ??\
P\
P\
P\
Pi
D.46)
4.6. Linear Elements
235
where uJ and pJ refer to the nodal components of elements j. The components of
these vectors are * and p\ where ? represents the node under consideration within
the element and / defines the component of displacements or tractions in the /
direction.
The functions ?1 are linear interpolation functions, such that,
*i = -i-J)
D.47)
If we consider N linear boundary elements the governing equation after
neglecting boundary forces for simplicity, becomes,
+
t {I ?*?<*? W = t {J *?drip' D.48)
Some of these integrals can be evaluated using numerical integration, others
- those with the singularity - can still be computed analytically. Once they are
all
found, all the elements contribution can be assembled in much the same way as
for the linear potential problem (section 2.5). A major difference with the line
ar
potential problem is that now we have two unknowns per node instead of one.
Corner Points
Another important difference with the linear potential problem is the type of c'
coefficients (equation D.48)) required in the elasticity solution at corners. Wh
ile
these coefficients in the linear potential case were associated with the value o
f the
solid angle at the corner, those for the elasticity problems are more complicate
d
to find.
For smooth boundaries the c' is simply a diagonal matrix with { on the diagonal.
When the point i is at a corner however (figure 4.6) the limit of the fundamenta
l
solution tractions, i.e.
p* dT} D.49)
? (?, J
Figure 4.6 Corner point
236
Chapter 4 Two Dimensional Elastostatics
gives a different result. For instance, for the two dimensional problems discuss
ed
{ diagonal matrix gives the following result,
here the above limit instead of a
-1 [A - v)(n + 02 - 0,) + sin 20, - sin 202, cos 202 - cos 20,
8n(l-v) Lcos 202 - cos 20,, 4A - v)(n + 62- 0,) + sin 202 - sin 20,
so that,
D.50)
It is much more complex to obtain a general expression for Ilk in three dimensio
ns
since the slope discontinuity may be of different types. In principle however, o
ne
may always do the integration over the corresponding portion of the spherical
surface.
The simplest way of computing the diagonal submatrices of the H matrix,
which includes the ? submatrices is using rigid body consideration as shown in
Chapter 3 for bounded domains or regions tending to infinity.
Corner
Nodes
(a) Use of discontinuous elements at corners
I
applied pressure
nodes
\
f element
Point of discontinuity
(b) Use of discontinuous elements to simulate a
sudden change in boundary conditions
element
Figure 4.7 Linear discontinuous elements
4.6 Linear Elements
Boundary Conditions, Discontinuous Elements
237
As discussed in section 3.5 different strategies can be followed to define bound
ary
conditions at corners or points of discontinuity. A simple way to solve this
problem is by introducing the concept of discontinuous elements (figure 4.7) whi
ch
implies moving the points inside the element when the number of unknowns at
the corner - or point of discontinuity - is more than one in each direction. Thi
s
allows the user to apply all types of boundary conditions in adjacent elements i
n
a rather simple fashion. Furthermore the approach can be used even in the presen
ce
of high stress concentration or singularities at the corner giving in this case
also
acceptable results when the discretization around the singularity is sufficientl
y fine.
Following the same approach as in the potential case the discontinuous element
formulation can be written as follows.
The displacements and tractions over a linear element are written in terms of
the extreme point values as given by equations D.45) and D.46) respectively. If
the two nodes are shifted from the two ends distances a and b, respectively (see
figure 2.11) equations D.45) and D.76) can be particularized for those two
points. For instance in the case of displacements
???)
?2?
?
?? ?
?2(?
?? ?
?2(? ) ?
?
^?( ? ??^?)}
D.51)
where ?0 = Ba/Z) - 1 and $? = 1 - Bb//)
Equation D.51) can be inverted and after substitution in D.45) yields the value
of u at any point on the element in terms of the nodal values
D.52)
where Q is the inverse of the matrix in equation 4.51, i.e.
Q =
l-a-b
l-b 0 -a
0
0
l-b 0 -a
-b 0
I-a 0
. 0
b 0
\-a.
D.53)
238
The same relation can be written for the tractions
'P\
D.54)
After discretizing the boundary into N elements equation D.48) can be written
with the only difference that for those elements which are discontinuous the
integrals over the elements take the form:
D.55)
j u*#Q,
r,when solving an elastic problem continuous and discontinuous elements can be
used together in the same mesh with the only difference that for the latter the
integrals in equation D.55) should be used instead of those in D.48). Numerical
integration is used when the collocation point is not on the integration element
.
Otherwise the integrals for the discontinuous elements are carried at analytical
ly
after subdividing the element in two parts at both sides of the node. Notice tha
t c'
matrix when the node is shifted towards the interior of the element is always th
at
of a smooth boundary point.
4.7 Quadratic Elements
Constant and linear boundary elements are well suited to solve many plane
elasticity problems including those with infinite as well as finite domains and
some
problems which present regions of stress concentration. Their main limitation is
that they can not represent properly curved geometries. Problems involving flexu
re
also require the use of higher order elements as their deformations are difficul
t to
model using linear elements (constant elements for this type of problem give ver
y
poor results).
4.7 Quadratic Elements
239
The simplest and more versatile type of curved boundary element is the
quadratic for which the displacements and tractions can be represented as (figur
e
4.8)
?::?
?? ? ?2 ? ?3 ?"
? </>! ? ?2 0 (/>???
I
D.56)
? 01 ? 02 ? 0!
Pi
?\
\?\)
The ?! are quadratic interpolation functions such that,
D.57)
D.58)
where I; is the dimensionless coordinate along the element (figure 4.7).
The geometry of the element can also be considered as quadratic and is
represented by the nodes coordinates and the same interpolation functions ?{ use
d
for displacements and traction components, i.e.
?
?2 ? ?3
0 ?2 0 ?
??
?'3}
V1
x2
D.59)
240
Chapter 4 Two Dimensional Elastostatics
Nodal value of u or q
Nodal value of u or q
Nodal value of u or q
3- V3)
0)
B)
-C)
? ,,, ? ,,
+ 1
Node
1
2
3
i
-1
0
1
1
0
0
?2
0
1
0
</>?
0
0
1
Figure 4.8 Quadratic element
The discretized boundary integral equations can be written as follows,
NE
>=1
NE
D.60)
where N? is the number of elements.
The integrals along dT need now to be transformed to the homogeneous
coordinate system ? as was done in section 2.7 for potential problems. Hence the
element integrals in D.60) become,
J ?*? dT = J p*<D|G| dt;
-l
D.61)
j" ?*? dT = J u*O|G|,
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
241
where the Jacobian \G\ is,
where dxjd^ and dx2/d? are easily obtained by differentiating D.59)in terms off.
4.8 Elastostatics Code using Quadratic Elements (ELQUABE)
In what follows a FORTRAN code for isotropic elasticity is described. The progra
m
has the same organization as the constant element code previously studied.
All integer variables in this code have the same meaning as for the constant
element program (ELCONBE). The same applies to all real arrays with the
exception of XM and YM which are not required now. Also FI and DFI have a
slightly different form. The dimension of FI is BN), N being the number of nodes
and that of DFI is CN) or what is the same FNE), where NE is the number of
elements. The prescribed boundary conditions are read in DFI (six per element).
FI is used as the right hand side vector for the solution of the system of equat
ions.
Vectors FI and DFI are reordered after solution in such a way that all values of
displacements are stored in FI and tractions in DFI.
The program allows for the traction values at both sides of the nodes connecting
two elements to be different. This is a similar treatment as was done for linear
and
quadratic potential elements (codes POLINBE and POQUABE, sections 2.6 and
2.9). For each particular xt or x2 direction, (i) when both tractions are prescr
ibed
as different at both sides of the node the displacement is the only unknown; (ii
)
when the displacement and one traction is prescribed one on each side the other
traction is the unknown and (iii) if only the displacement is prescribed, one va
lue
of the traction is the only unknown and this will be the same at both sides of t
he
node.
In problems with only one region the case of having more unknowns than
boundary conditions at a corner point seldom occurs and does not present
difficulties. If the displacement is known along the two boundary elements that
converge into a corner, the derivatives along these two directions are known as
well, thus strains, stresses and tractions are known. One only needs to prescrib
e
two variables along each direction and leave the third (displacement, traction
before the node or after the node) as a check that it coincides with its known
value. Only in cases of discontinuities of the stress tensor at certain corners
one
needs to implement the problem of corners in a different way. The same happens
for non-homogeneous media. An accurate and simple procedure is the use of
242
Chapter 4 Two Dimensional Elastostatics
discontinuous elements, i.e. elements that have one of the node shifted inside t
he
element in such a way that a different node will be created for each element
converging into the corner.
1. Main Program
This program follows the same structure as the constant elasticity code in secti
on
4.5.
Its present dimensions allow for 50 elements and 100 nodes.
The listing is as follows:
PROGRAM ELQUABE
?
? PROGRAM 36
?
?
? THIS PROGRAM SOLVES TWO DIMENSIONAL (EL)ASTIC PROBLEMS
? USING (QUA)DRATIC (B)OUNDARY (E)LEMENTS
?
CHARACTER*10 FILEIN,FILEOUT
?
COMMON/MATG/ GA00,150)
COMMON/MATH/ HA00,100)
COMMON N,L,GE,XNU,INP,IPR
DIMENSION XE1),YE1)
DIMENSION DFIA50),FIA00),KODEA50)
DIMENSION CXB0),CYB0),SSOLF0),DSOLD0)
?
? SET MAXIMUN DIMENSION OF THE SYSTEM OF EQUATIONS (NX)
? NX= 2 MAXIMUN NUMBER OF NODES= 4*MAXIMUN NUMBER OF ELEMENTS
? NX1= 3*MAXIMUN NUMBER OF NODES= 6*MAXIMUN NUMBER OF ELEMENTS
?
NX=100
NX1=15O
?
? ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES
?
INP=5
IPR=6
?
? READ NAMES AND OPEN FILES FOR INPUT AND OUTPUT
?
WRITE(*,' (A) ') ' NAME OF INPUT FILE (MAX. 10 CHART.)'
READ(*,' (A) ')FILEIN
OPEN(INP,FILE=FILEIN,STATUS='OLD')
WRITEf*,' (A) ') ' NAME OF OUTPUT FILE (MAX. 10 CHART.)'
READ(*,' (A) ')FILEOUT
OPEN(IPR,FILE=FILEOUT,STATUS'NEW)
?
? READ DATA
?
CALL INPUTEQ(CX,CY,X,Y,KODE,DFI)
?
? COMPUTE G AND H MATRICES AND FORM SYSTEM (A X = F)
?
CALL GHMATEQ(X,Y,G,H,FI,DFI,KODE,NX,NX1)
?
? SOLVE SYSTEM OF EQUATIONS
?
NN=2*N
CALL SLNPD(H,FI,D,NN,NX)
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
243
?
? COMPUTE STRESS AND DISPLACEMENT VALUES AT INTERNAL POINTS.
?
CALL INTEREQ(FI,DFI,K0DE,CX,CY,X,Y,SSOL,DSOL)
?
? PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS
?
CALL OUTPTEO.(X,Y,FI,DFI,CX,CY,SSOL,DSOL)
?
? CLOSE INPUT AND OUTPUT FILES
?
CLOSE (INP)
CLOSE (IPR)
STOP
END
2. Routine INPUTEQ
This subroutine reads all the input required by the program and requests a file
from
the user containing the following lines of free format input;
(i) Title Line Contains the title of the problem.
(ii) Basic Parameter Lines Contains the number of elements, number of
internal points, the shear modulus and the Poisson ratio (or fictitious
Poisson's ratio for plane stress problems for which v' = v/(l + v)).
(iii) Boundary Nodes Coordinates Lines The coordinates are read countercounterclockwise for external boundaries and clockwise for internal ones.
(iv) Boundary Conditions Lines As many lines as boundary elements. Six values
of KODE and the known variables are read for each element, corresponding
to the three nodes and two directions per node. In this way, a value of a
traction may be prescribed for a node as part of one element and a different
value as part of the other element. The displacement however must be unique
for any node. The variables are displacement if KODE = 0 or traction if
KODE=1.
(v) Internal Points Coordinate Lines The xtx2 coordinates for each point are
read in free format. There will be one or more lines, if necessary, i.e. if
internal points have been defined.
This subroutine first prints the name of the run and the basic parameters. Then
the coordinates of the nodes and the boundary conditions given by element, with
codes and prescribed values, are printed. The internal point coordinates will be
printed in the subroutine OUTPTEQ.
The FORTRAN listing of INPUTEQ is as follows:
?
SUBROUTINE INPUTEQ(CX,CY,X,Y,KODE,DFI) ,
?
?
PROGRAM 37
?
?
NE= NUMBER OF BOUNDARY ELEMENTS
?
L = NUMBER OF INTERNAL POINTS
?
GE= SHEAR MODULUS
244
Chapter 4 Two Dimensional Elastostatics
? XNU= POISSON RATIO
?
CHARACTER*80 TITLE
COMMON N,L,GE,XNU,INP,IPR
DIMENSION KODE(l),XA),YA),CXA),CYA),DFIA)
WRITE(IPR,100)
100 FORMAT(' ',79('*'))
?
? READ JOB TITLE
?
READ(INP,' (A)') TITLE
WRITE (IPR, ' (A) ') TITLE
?
? READ NUMBER OF BOUNDARY ELEMENTS AND INTERNAL POINTS;
? READ MATERIAL PROPERTIES
?
READ(INP,*)NE,L,GE,XNU
WRITE (IPR, 210) NE, L,GE, XNU
210 FORMAT(//2X, 'DATA'/2X, 'NUMBER OF BOUNDARY ELEMENTS=',
1I3/2X, 'NUMBER OF INTERNAL POINTS=', I3/2X, 'SHEAR MODULUS=',
2E14.7/2X,'POISSON RATIO=',E14.7)
N=2*NE
?
? READ BOUNDARY NODES COORDINATES
?
N=2*NE
WRITE(IPR,500)
READ(INP,*) (X(I),Y(I),I=1,N)
DO 10 1=1,N
10 WRITE(IPR,240) I,X(I),Y(I)
500 FORMAT(//2X,'BOUNDARY NODES COORDINATES'///4X,
l'NODE',10X,'X',18X,'Y'/)
240 FORMATEX,I3,2EX,E14.7))
?
? READ BOUNDARY CONDITIONS IN DFI(I) VECTOR, IF KODE(I)=0
? THE DFI(I) VALUE IS A KNOWN DISPLACEMENT; IF KODE(I)=1 THE
? DFI(I) VALUE IS A KNOWN TRACTION.
? SIX BOUNDARY CONDITIONS ARE READ PER ELEMENT.
? NODES BETWEEN TWO ELEMENTS MAY HAVE TWO DIFFERENT VALUES
? OF THE TRACTION BUT ONLY ONE VALUE OF THE DISPLACEMENT.
?
WRITE(IPR,800)
800 FORMAT(//2X,'BOUNDARY CONDITIONS'//35X,'PRESCRIBED VALUES' /
17X,' FIRST NODE
',3X,' SECOND NODE
',3X,
2'
THIRD NODE
'/
31X,'ELE',3X,'X DIR.',2X,'C',3X,'Y DIR. ', 2X,'C , 3X,
4'X DIR.',2X,'C',3X,'Y DIR. ' ,2X,'C , 3X,'X DIR.',
52X,'C',3X,'Y DIR.',2X,'C'/)
DO 20 1=1,NE
READ(INP,*) (K0DEF*I-6+J),DFIF*I-6+J),J=1,6)
20 WRITE(IPR,950I,(DFIF*I-6+J),KODEF*I-6+J),J=1,6)
950 FORMATAX,I3,6(F9.3,2X,I1))
?
? READ INTERNAL POINTS COORDINATES
?
IF(L.EQ.O) GO TO 30
READ(INP,*)
30 RETURN
END
3. Routine GHMATEQ
This subroutine computes the G and H system matrices by calling routines
EXTINEQ and LOCINEQ.
EXTINEQ: Computes the GW and HW Bx6) submatrices which relates a
collocation point with an element as defined by its three nodes.
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
245
LOCINEQ: Computes the GW Bx6) submatrix for the case when the
collocation point is one of the nodes within the element under
consideration (i.e. the singularity is in the same element). Notice
that the corresponding H W B x 6) is computed using EXTINEQ
except for that part that relates a node with itself which is computed
using rigid body considerations, which results in adding row
coefficients.
The resulting GW and HWsubmatrices are assembled in the G and H system
matrices. Matrix G is now rectangular since each extreme node of an element may
have different tractions, i.e. one 'before' and another 'after' the node.
Once the matrices H and G are assembled, the system of equations needs to
be reordered in accordance with the boundary conditions to form
AX = F
where X is a B??) vector of unknowns, N being the number of nodes; A is a
BN x 2N) matrix whose columns are a combination of columns of H or G
depending on the boundary conditions or of two consecutive columns of G when
the unknown is the unique value of the tractions at both sides of the extreme no
de
of an element; F is a known vector computed by multiplying the prescribed
boundary conditions by the corresponding row terms of G or H.
At the end of the subroutine GHMATEQ and after rearranging ? contains
the matrix A, and FI the F vector.
The FORTRAN listing of GHMATEQ is as follows,
?
SUBROUTINE GHMATEQ ( X, Y, G, H, FI, DFI, KODE, NX, NX1)
?
? PROGRAM 38
?
? THIS SUBROUTINE COMPUTES THE G AND H MATRICES AND FORMS
? THE SYSTEM A X = F READY TO BE SOLVED
? ? IS A SQUARE MATRIX D*NE,4*NE) ; G IS RECTANGULAR D*NE,6*NE)
?
DIMENSION X(l),YA),G(NX,NX1),H(NX,NX),HWB,6),GWB,6)
DIMENSION FIA),DFIA),KODEA)
COMMON N,L,GE,XNU,INP,IPR
NN=2*N
NE=N/2
DO 20 1=1,NN
DO 11 J=1,NN
11
H(I,J)=0.
DO 12 J=1,3*N
12
G(I,J)=0.
20
CONTINUE
X(N+1)=XA)
Y(N+1)=YA)
?
? COMPUTE THE GW AND HW MATRICES FOR EACH COLLOCATION
IF(KODE(J-4)) 115,115,113
113
DO 114 K=1,NN
CH=H(K,J-4)
H(K,J-4)=-G(K,6*I-6+J)*GE
114
G(K,6*I-6+J)=-CH
GO TO 170
115
DO 116 K=1,NN
H(K,J-4)=H(K,J-4)-G(K,6*I-6+J)*GE
116
G(K,6*I-6+J)=0.
GO TO 170
125 IF(I.EQ.l .OR. J.GT.2 .OR. KODEF*I-8+J).EQ.1) GO TO 130
DO 129 K=1,NN
H(K,4*I-4+J)=H(K,4*I-4+J)-G(K,6*I-6+J)*GE
129
G(K,6*I-6+J)=0.
GO TO 170
130
DO 132 K=1,NN
CH=H(K,4*I-4+J)
H(K,4*I-4+J)=-G(K,6*I-6+J)*GE
132 G(K,6*I-6+J)=-CH
170 CONTINUE
180 CONTINUE
?
? FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN FI
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
247
?
DO 190 I=1,NN
FI(I)= 0.
DO 185 J=1,6*NE
185 FI(I)=FI(I)+G(I,J)*DFI(J)
190 CONTINUE
RETURN
END
4. Routine EXTINEQ
This subroutine computes using numerical integration, the B x 6) submatrices
GW and HW that corresponds to an element when the collocation point is at a
node other than any of those 3 in the element. The coordinates of the collocatio
n
points are XP and YP. The integrals are of the type D.61), i.e.
HW = J ?*? dt, = j p*0|G| d?
D.63)
r,
-l
+ i
GW= $ u*?d?= $ u*?\G\dZ D.64)
They can be expanded as follows,
+ 1??* ?* 1??1 0 ?2 ? ?3 ?
HW= J ?'1 P4
|G|^
D.65)
~i LP21 P22JL 0 ^>j ? ?2 0 '/'sJ
?1 ? ?2 0 1 _
The Jacobians are calculated by taking derivatives of the expressions for the xy
and x2 coordinates, in the same way that was done in section 2.9 for potential
problems using quadratic elements.
The FORTRAN listing is as follows:
SUBROUTINE EXTINEQ(XP,YP,X1,Y1,X2,Y2,X3,Y3,HW,GW)
?
? PROGRAM 39
?
? THIS SUBRUOTINE COMPUTES THE HW AND GW MATRICES
? THAT RELATE A NODE (XP,YP) WITH A BOUNDARY
? ELEMENT USING GAUSS QUADRATURE.
?
? RA
- RADIUS
? RD1,RD2,RDN = RADIUS DERIVATIVES
? ???1,???2 = COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
? XCO,YCO = INTEGRATION POINT ALONG THE ELEMENT
? XJA = JACOBIAN
?
COMMON N,L,GE,XNU,INP,IPR
DIMENSION GWB,6),HWB,6)
DIMENSION GIA0),OMEA0)
DATA GI/O.9739065285,-0.9739065285,0.8650633666,-0.8650633666
248
Chapter 4 Two Dimensional Elastostatics
6,0.6794095683,-0.6794095682,0.4333953941,-0.4333953941,
60.1488743389,-0.1488743389/
DATA OME/0.0666713443,0.0666713443,0.1494513491,0.1494513491
6,0.2190863625,0.2190863625,0.2692667193,0.2692667193,
60.2955242247,0.2955242247/
DO 30 1=1,2
DO 20 J=l,6
HW(I,J)=0.
20 GW(I,J)=0.
30 CONTINUE
A=X3-2*X2+X1
B=(X3-Xl)/2
C=Y3-2*Y2+Y1
D=(Y3-Yl)/2
DE=4*3.141592*A-XNU)
DO 40 1=1,10
?
? COMPUTE THE VALUES OF THE SHAPE FUNCTIONS AT THE
? INTEGRATION POINTS
?
Fl<=GI(I)*(GI(I)-l)*0.5
F2=1.-GI(I)**2
F3=GI(I)*(GI(I)+l)*0.5
?
? COMPUTE GEOMETRICAL PROPERTIES AT THE INTEGRATION POINTS
?
XCO=X1*F1+X2 *F2+X3 *F3
YCO=Y1*F1+Y2 F2+Y3 *F3
XJA=SQRT((GI(I)*A+B)**2+(GI(I)*C+D)**2)
ETA1=(GI(I)*C+D)/XJA
ETA2=-(GI(I)*A+B)/XJA
RA=SQRT((XP-XCO)**2+(YP-YC0)**2)
RD1=(XCO-XP)/RA
RD2=(YCO-YP)/RA
RDN=RD1*ETA1+RD2 *ETA2
?
? COMPUTE GW AND HW MATRICES
?
GW(l,l)=GW(l,l)+(C-4*XNU)*ALOG(l./RA)+RDl**2)*OME(I)*XJA*Fl/
1B*DE*GE)
GWA,2)=GWA,2)+RD1*RD2*OME(I)*XJA*F1/B*DE*GE)
GWB,1)=GWA,2)
GWB,2)=GWB,2)+(C-4*XNU)*ALOG(l./RA)+RD2**2)*OME(I)*XJA*Fl/
1B*DE*GE)
HWA,1)=HWA,1)-RDN*(A-2*XNU)+2*RD1**2)/(RA*DE)*OME(I)*XJA*F1
HWA,2)=HWA,2)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA1*RD2-ETA2*RD1))*
1OME(I)*XJA*F1/(RA*DE)
HWB,1)=HWB,1)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA2*RD1-ETA1*RD2))*
1OME(I)*XJA*F1/(RA*DE)
HWB,2)=HWB,2)-RDN*(A-2*XNU)+2*RD2**2)*OME(I)*XJA*F1/(RA*DE)
GW(l,3)=GW(l,3)+(C-4*XNU)*ALOG(l./RA)+RDl**2)*OME(I)*XJA*F2/
1B*DE*GE)
GW A, 4) =GW A,4) +RD1*RD2*OME (I) *XJA*F2/ B*DE*GE)
GWB,3)=GWA,4)
GWB,4)=GWB,4)+(C-4*XNU)*ALOG(l./RA)+RD2**2)*OME(I)*XJA*F2/
1B*DE*GE)
HWA,3)=HWA,3)-RDN*(A-2 *XNU)+2 *RD1**2)/(RA*DE)*OME(I)*XJA*F2
HWA,4)=HWA,4)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA1*RD2-ETA2*RD1))*
1OME(I)*XJA*F2/(RA*DE)
HWB,3)=HWB,3)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA2*RD1-ETA1*RD2))*
1OME(I)*XJA*F2/(RA*DE)
HWB,4)=HWB,4)-RDN*(A-2*XNU)+2*RD2**2)*OME(I)*XJA*F2/(RA*DE)
GW(l,5)=GW(l,5)+(C-4*XNU)*ALOG(l./RA)+RDl**2)*OME(I)*XJA*F3/
1B*DE*GE)
GWA,6)=GWA,6)+RD1*RD2*OME(I)*XJA*F3/B*DE*GE)
GWB,5)=GWA,6)
GWB,6)=GWB,6)+(C-4*XNU)*ALOG(l./RA)+RD2**2)*OME(I)*XJA*F3/
1B*DE*GE)
HWA,5)=HWA,5)-RDN*(A-2*XNU)+2*RD1**2)/(RA*DE)*OME(I)*XJA*F3
HWA,6)=HMA,6)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA1*RD2-ETA2*RD1))*
1OME(I)*XJA*F3/(RA*DE)
HWB,5)=HWB,5)-(RDN*2*RD1*RD2+A-2*XNU)*(ETA2*RD1-ETA1*RD2))*
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
249
10ME(I)*XJA*F3/(RA*DE)
40 HWB,6)=HWB,6)-RDN*(A-2*XNU)+2*RD2**2)*OME(I)*XJA*F3/(RA*DE)
RETURN
END
5. Routine LOCINEQ
This subroutine computes using standard Gauss quadrature and a special
quadrature formula, the B x 6) submatrix GW when the collocation point is one
of the nodes within the element under consideration.
The integration process is completely analogous to that shown for potential
problems in section 2.8. The integrals are split into two parts: one without
singularity which is integrated by means of the standard formula; and the other,
the loarithmic part, that is integrated using a special quadrature formula. More
details of the integration process are given in section 2.8. The only difference
is
now that kernels to be integrated have some more terms that in the case of poten
tial
problems. Also GW is now B x 6) instead of C).
The listing for this subroutine is as follows:
SUBROUTINE LOCINEQ(XG1,YG1,XG2, YG2,XG3,YG3,GW,NODO)
?
? PROGRAM 40
?
? THIS SUBROUTINE COMPUTES THE GW MATRIX WHEN THE COLLOCATION
? POINT IS ONE OF THE NODES OF THE INTEGRATION ELEMENT.
? THE COEFFICIENTS ARE COMPUTED BY NUMERICAL INTEGRATION:
? THE NON SINGULAR PART IS COMPUTED USING STANDARD GAUSS QUADRATURE,
? THE LOGARITHMIC PART IS COMPUTED USING A SPECIAL QUADRATURE FORMULA.
?
COMMON N,L,GE,XNU,INP,IPR
DIMENSION GIA0),OMEA0),GILA0),OMELA0),GWB,6),RB)
?
? DATA FOR THE GAUSS QUADRATURE
?
DATA GI/0.9739065285,-0.9739065285,0.8650633666,-0.8650633666
,0.6794095682,-0.6794095682,0.4333953941,-0.433 3953941,
0.1488743389,-0.1488743389/
DATA OME/0.0666713443, 0.0666713443,0.1494513491,0.1494513491
,0.2190863625,0.2190863625,0.2692667193,0.2692667193,
gO.2955242247,0.2955242247/
?
? DATA FOR THE SPECIAL QUADRATURE
?
DATA GIL/0.0090426309,0.0539712662,0.1353118246,0.2470524162
,0.3802125396,0.5237923179,0.6657752055,0.7941904160,
0.8981610912,0.9688479887/
DATA OMEL/0.1209551319, 0.1863635425,0.1956608732,0.1735771421
,0.1356956729,0.0936467585,0.0557877273,0.0271598109,
0.0095151826,0.0016381576/
?
? SET LOCAL COORDINATES SYSTEM
?
GOTOA,2,3),NODO
1 X3=XG3-XG1
Y3=YG3-YG1
X2=XG2-XG1
Y2=YG2-YG1
Al=(X3-2*X2)*0.5
B1=X2
A2=(Y3-2*Y2)*0.5
B2=Y2
GO TO 4
250
Chapter 4 Two Dimensional Elastostatics
2
X3=XG3-XG2
Y3=YG3-YG2
X1=XG1-XG2
Y1=YG1-YG2
A1=X1+X3
B1=X3-X1
A2=Y1+Y3
B2=Y3-Y1
GO TO 4
3
X2=XG2-XG3
Y2=YG2-YG3
X1=XG1-XG3
Y1=YG1-YG3
Al=(Xl-2*X2)*0.5
Bl=-X2
A2=(Yl-2*Y2)*0.5
B2=-Y2
4
CONTINUE
?
DO 10 1=1,2
DO 10 J=l,6
10 GW(I,J)=0.
A=A1**2 + A2**2
B=2*(A1*B1 + A2*B2)
C=B1**2 + B2**2
CONT1=C-4*XNU)/(8.*3.1415926*GE*A-XNU))
CONT2 =CONT1/C-4*XNU)
DO 250 1=1,10
T1=((A1*GI(I)+B1)**2)/((A2*GI(I)+B2)**2 + (A1*GI(I)+B1)**2)
T2=((A2*GI(I)+B2)**2)/((A2*GI(I)+B2)**2 + (A1*GI(I)+B1)**2)
T3=(A1*GI(I)+B1)*(A2*GI(I)+B2)/((A1*GI(I)+B1)**2+(A2*GI(I)+B2)**2)
?
1XB*I+1),YB*I+1),HW,GW)
CALL SIGMAEQ(CX(K),CY(K),XB*I-1),YB*1-1),XB*I),YB*I),
1XB*I+1),YB*I+1),D11,D12,D22,S11,S12,S22)
DO 220 J=l,6
IJ4=4*I-4+J
IF(IJ4.GT.D*NE)) IJ4=J-4
SSOLC*K-2)=SSOLC*K-2)+Dll(J)*DFIF*I-6+J)-Sll(J)*FI(IJ4)
SSOLC*K-1)=SSOLC*K-1)+D12(J)*DFIF*I-6+J)-S12(J)*FI(IJ4)
SSOLC*K)=SSOLC*K)+D22(J)*DFIF*I-6+J)-S22(J)*FI(IJ4)
DSOLB*K-l)=DSOLB*K-l)+GW(l,J)*DFIF*I-6+J)-HW(l,J)*FI(IJ4)
220 DSOLB*K)=DSOLB*K)+GWB,J)*DFIF*I-6+J)-HWB,J)*FI(IJ4)
230 CONTINUE
240 CONTINUE
50 RETURN
END
7. Routine SIGMAEQ
The integrals of the S and D coefficients multiplied by the shape functions in
equation D.63) are evaluated using Gaussian quadrature. The procedure is similar
to that used in routine EXTINEQ, the only difference being that now the
expressions of D and S given by equation C.110) and C.111) are computed at the
integration points instead of the values of u* and p* as in EXTINEQ.
The FORTRAN listing of SIGMAEQ is as follows,
Q
,
SUBROUTINE SIGMAEQ(XP,YP,XI,Yl,X2,Y2,X3,Y3,Dll,D12,D22,Sll,S12
*,S22)
?
? PROGRAM 42
?
? THIS SUBROUTINE COMPUTES THE VALUES OF THE S AND D MATRICES
? USING GAUSS QUADRATURE IN ORDER TO COMPUTE THE STRESS
? AT ANY INTERNAL POINT.
?
?
?
?
?
?
RA
RD1,
???1
???,
XJA
RD2,RON
,???2
YCO
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
253
RADIUS
RADIOS DERIVATIVES
COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT
INTEGRATION POINT ALONG THE ELEMENT
JACOBIAN
COMMON N,L,GE,XNU,INP,IPR
DIMENSION DllF),D12F),D22F),S11F),S12F),S22F)
DIMENSION GIA0),OMEA0)
DATA GI/O.9739065285,-0.9739065285,0.8650633666,-0.8650633666
,0.6794095683,-0.6794095682,0.4333953941,-0.4333953941,
??.1488743389,-0.1488743389/
DATA OME/0.0666713443, 0.0666713443, 0.1494513491, 0.1494513491
,0.2190863625,0.2190863625,0.2692667193,0.2692667193,
0.2955242247,0.2955242247/
DO 20 J=l,6
D11(J)=O.
D12(J)=0.
D22(J)=0.
S11(J)=O.
S12(J)=0.
20 S22(J)=0.
?
FA=1-4*XNU
AL=1-2*XNU
A=X3-2*X2+X1
B=(X3-Xl)/2
C=Y3-2*Y2+Y1
D=(Y3-Yl)/2
DE=4*3.141592*A-XNU)
DO 40 1=1,10
?
? COMPUTE VALUES OF THE SHAPE FUNCTIONS AT THE INTEGRATION POINTS
?
Fl=GI(I)*(GI(I)-l)*0.5
F2=1.-GI(I)**2
F3=GI(I)*(GI(I)+l)*0.5
?
? COMPUTE GEOMETRICAL PARAMETERS
?
XCO=X1*F1+X2*F2+X3*F3
YCO=Y1*F1+Y2*F2+Y3*F3
XJA=SQRT((GI(I)*A+B)**2+(GI(I)*C+D)**2)
ETA1=(GI(I)*C+D)/XJA
ETA2=-(GI(I)*A+B)/XJA
RA=SQRT((XP-XCO)**2+(YP-YCO)**2)
RD1=(XCO-XP)/RA
RD2=(YCO-YP)/RA
RDN=RD1*ETA1+RD2*ETA2
?
? COMPUTE D AND S COEFFICIENTS
?
D11A)=D11A) + (AL*RD1+2*RD1**3)*OME(I)*XJA*F1/(DE*RA)
D11B)=D11B) + B*RD1**2*RD2-AL*RD2)*OME(I)*XJA*F1/(DE*RA)
D11C)=D11C) + (AL*RD1+2*RD1**3)*OME(I)*XJA*F2/(DE*RA)
D11D)=D11D) + B*RD1**2*RD2-AL*RD2)*OME(I)*XJA*F2/(DE*RA)
D11E)=D11E) + (AL*RD1+2*RD1**3)*OME(I)*XJA*F3/(DE*RA)
D11F)=D11F) + B*RD1**2*RD2-AL*RD2)*OME(I)*XJA*F3/(DE*RA)
D12A)=D12A) + (AL*RD2+2*RD1**2*RD2) *F1/(DE*RA) *OME(I) *XJA
L*RD1+2*RD1*RD2**2)*F1/(DE*RA)*OME(I)*XJA
L*RD2+2*RD1**2*RD2)*F2/(DE*RA)*OME(I)*XJA
()() (L*RD1+2*RD1*RD2**2)*F2/(DE*RA)*OME(I)*XJA
D12E)=D12E)+(AL*RD2+2*RD1**2*RD2)*F3/(DE*RA)*OME(I)*XJA
D12F)=D12F) + (AL*RD1+2*RD1*RD2**2)*F3/(DE*RA)*OME(I)*XJA
D22A)=D22A) + B*RD1*RD2**2-AL*RD1)*F1/(DE*RA)*OME(I)*XJA
D22B)=D22B)+(AL*RD2+2*RD2**3) *F1/ (DE*RA) *OME(I) *XJA
D22C)=D22C) + B*RD1*RD2**2-AL*RD1)*F2/(DE*RA)*OME(I)*XJA
D22 D) =D22 D) + (AL*RD2+2*RD2**3) *F2/ (DE*RA) *OME(I) *XJA
D22 E) -D22 E) + B*RDl*RD2**2-AL*RDl) *F3/(DE*RA) *OME(I) *XJA
D22F)=D22F) + (AL*RD2+2*RD2**3)*F3/(DE*RA)*OME(I)*XJA
S11A)=S11A) + B*RDN*(AL*RD1+XNU*2*RD1-4*RD1**3)+4*XNU*ETA1
1*RD1**2+AL*B*ETA1*RD1**2+2*ETA1)-FA*ETA1)*2*GE*F1/(DE*RA**2)*
2OME(I)*XJA
254
Chapter 4 Two Dimensional Elastostatics
S11B)=S11B) + B*RDN*(AL*RD2-4*RD1**2*RD2)+4*XNU*ETA1*RD1*RD2+
1AL*2*ETA2*RD1**2-FA*ETA2) *2*GE*F1/ (DE*RA**2) *OME(I) *XJA
S11C)=S11C)+B*RDN*(AL*RD1+XNU*2*RD1-4*RD1**3)+4*XNU*ETA1
1*RD1**2+AL* B*ETA1*RD1**2+2*ETA1) -FA*ETA1) *2*GE*F2/ (DE*RA**2) *
2OME(I)*XJA
S11D)=S11D) + B*RDN*(AL*RD2-4*RD1**2*RD2)+4*XNU*ETA1*RD1*RD2+
1AL*2*ETA2*RD1**2-FA*ETA2) *2*GE*F2/ (DE*RA**2) *OME(I) *XJA
S11E)=S11E) + B*RDN*(AL*RD1+XNU*2*RD1-4*RD1**3)+4*XNU*ETA1
1*RD1**2+AL*B*ETA1*RD1**2+2*ETA1)-FA*ETA1)*2*GE*F3/(DE*RA**2)*
2OME(I)*XJA
S11F)=S11F) + B*RDN*(AL*RD2-4*RD1**2*RD2)+4*XNU*ETA1*RD1*RD2+
1AL*2*ETA2*RD1**2-FA*ETA2)*2*GE*F3/(DE*RA**2)*OME(I)*XJA
S12A)=S12A)+B*RDN*(XNU*RD2-4*RD1**2*RD2)+2*XNU*(ETA1*RD2*
1RD1+ETA2*RD1**2)+AL*B*ETA1*RD1*RD2+ETA2))*2*GE*F1/
2(DE*RA**2)*OME(I)*XJA
S12B)=S12B)+B*RDN*(XNU*RD1-4*RD1*RD2**2)+2*XNU*(ETA1*RD2**2
1+ETA2*RD1*RD2)+AL*B*ETA2*RD1*RD2+ETA1)) *2*GE*F1/
2(DE*RA**2)*OME(I)*XJA
S12C)=S12C) + B*RDN*(XNU*RD2-4*RD1**2*RD2)+2*XNU*(ETA1*RD2*
1RD1+ETA2*RD1**2)+AL*B*ETA1*RD1*RD2+ETA2))*2*GE*F2/
2(DE*RA**2)*OME(I)*XJA
S12D)=S12D)+B*RDN*(XNU*RD1-4*RD1*RD2**2)+2*XNU*(ETA1*RD2**2
1+ETA2*RD1*RD2)+AL*B*ETA2*RD1*RD2+ETA1))*2*GE*F2/
2(DE*RA**2)*OME(I)*XJA
S12E)=S12E)+B*RDN*(XNU*RD2-4*RD1**2*RD2)+2*XNU*(ETA1*RD2*
1RD1+ETA2*RD1**2)+AL*B*ETA1*RD1*RD2+ETA2))*2*GE*F3/
2(DE*RA**2)*OME(I)*XJA
S12F)=S12F)+B*RDN*(XNU*RD1-4*RD1*RD2**2)+2*XNU*(ETA1*RD2**2
1+ETA2*RD1*RD2)+AL*B*ETA2*RD1*RD2+ETA1))*2*GE*F3/
2(DE*RA**2)*OME(I)*XJA
S22A)=S22A)+B*RDN*(AL*RD1-4*RD1*RD2**2)+4*XNU*ETA2*RD1*RD2+
1AL*2*ETA1*RD2**2-FA*ETA1)*2*GE*F1/(DE*RA**2)*OME(I)*XJA
S22B)=S22B)+B*RDN*(AL*RD2+2*XNU*RD2-4*RD2**3)+4*XNU*ETA2
1*RD2**2+AL*B*ETA2*RD2**2+2*ETA2)-FA*ETA2)*2*GE*F1/
2(DE*RA**2)*OME(I)*XJA
S22C)=S22C)+B*RDN*(AL*RD1-4*RD1*RD2**2)+4*XNU*ETA2*RD1*RD2+
1AL*2*ETA1*RD2**2-FA*ETA1)*2*GE*F2/(DE*RA**2)*OME(I)*XJA
S22D)=S22D)+B*RDK*(AL*RD2+2*XNU*RD2-4*RD2**3)+4*XNU*ETA2
1*RD2**2+AL*B*ETA2*RD2**2+2*ETA2)-FA*ETA2)*2*GE*F2/
2(DE*RA**2)*OME(I)*XJA
S22E)=S22E)+B*RDN*(AL*RD1-4*RD1*RD2**2)+4*XNU*ETA2*RD1*RD2+
1AL*2*ETA1*RD2**2-FA*ETA1)**2*GE*F3/(DE*RA**2)*OME(I)*XJA
40 S22F)=S22F)+B*RDN*(AL*RD2+2*XNU*RD2-4*RD2**3)+4*XNU*ETA2
1*RD2**2+AL*B*ETA2*RD2**2+2*ETA2)-FA*ETA2)*2*GE*F3/
2(DE*RA**2)*OME(I)*XJA
RETURN
END
8. Routine OUTPTEQ
This subroutine prints the results in the following order.
(i) Displacements at boundary nodes.
(ii) Tractions at boundary nodes (tractions 'before' and 'after' each node are
printed).
(iii) Internal point displacements,
(iv) Internal point stresses.
The listing is as follows:
?
SUBROUTINE OUTPTEQ(X,Y,FI,DFI,CX,CY,SSOL,DSOL)
?
? PROGRAM 43
?
-2.87295 -0.76985,
-2.9745 0.
-2.87295 0.76985
-2.57585 1.4872
-2.1032 2.1032
-1.4872 2.57585
-0.76985 2.87295
0. 2.9743
256
Chapter 4 Two Dimensional Elastostatics
0.76985
1.4872
2.1032
2.57585
2.87295
2.9745
2.87295
2.57585
2.1032
1.4872
0.76985
0.
1 -25.
1 -70.
1 -96.
1 -96.
1 -70.
1 -25.
1 25.88
1 70.71
1 96.59
1 96.59
1 70.71
1 25.88
4. 0. 2
2.87295
2.57575
2.1032
1.4872
0.76985
0.
-0.76985
-1.4872
-2.1032
-2.57585
-2.87295
-2.9743
88 1 -96.59
71 1 -70.71
59 1 -25.88
59 1 25.88
71 1 70.71
88 1 96.59
1 96.59 1
1 70.71 1
1 25.88 1
1 -25.88 1
1 -70.71 1
1 -96.59 0
1
1
1
1
1
0
50.
86.
100
86.
50.
0
-50.
-86.
-100
-86.
-50.
0.
1
6 1
. 0
6 1
1
. 1
.82843 2.82843 -4. 0
1
6 1
. 1
6 1
1
1
86.
50.
0.
-50.
-86.
-86
-50
0.
50
86
100
,6 1
, 1
1
1
6 1
-100.
. 6.
0.
.6
.
.6
. 1
70
96
96
70
25
1
10
1
1
1
1
1
25
.71
.59
.59
.71
.88
-25
. 0
-70
-96
-96
-70
-25
.88
1
1
1
1
1
.88
.71
.59
.59
.71
.88
1
70.
25.
-25.
-70.
-96.
1
1
1
1
1
96
71
88
88
71
59
1 -96
-70
-25
25
70
96
.59
.59
.71
.88
.88
.71
.59
The output given below also presents symmetry of results, same absolute values
(with different sign) for hoop and radial stresses at internal points and expone
ntial
decay of both displacements and stresses with distance from the cavity.
The comparison is as follows.
Table 4.2 12 Quadratic Elements Discretization
Distance from centre
of cavity
4
6
10
Elasticity
theory
-56.25
-25.00
-9.00
Constant
Stress
-57.234
-25.295
-9.106
elements
% Error
1.75%
1.18%
1.18%
Quadratic
Stress
-55.275
-24.565
-8.844
elements
% Error
1.73%
1.74%
1.73%
(Notice that only internal points with comparatively large stresses are taken in
to
consideration as the stresses are too low for points further away from the cavit
y
to give a meaningful error estimate.)
It is interesting to see that in this case the use of quadratic elements does no
t
appreciably alter the error in the results on the contrary, they are overall les
s
accurate for the same number of degrees of freedom. This is due to the problem
being rather of a special character. As we will see in what follows, although co
nstant
elements behave well in problems such as this they give poor results when used
to model problems with bending.
CIRCULAR CAVITY (OUTPUT)
*****************************************************************************
CIRCULAR CAVITY UKDER INTERNAL PRESSURE A2 QUADRATIC ELEMENTS)
DATA
NUMBER OF BOUNDARY ELEMENTS= 12
NUMBER OF INTERNAL POINTS= 5
SHEAR MODULUS= ?.9450000?+05
POISSON RATI0= 0.1000000E+00
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
257
0.2872950E+01
0.2974300E+01
0.2872950E+01
0.2575750E+01
0.2103200E+01
0.1487200E+01
0.7698500E+00
0.0000000E+00
-0.7698500E+00
-0.1487200E+01
-0.2103200E+01
-0.2575850E+01
-0.2872950E+01
-0.2974300E+01
BOUNDARY CONDITIONS
ELE
1
2
3
4
5
6
7
8
9
10
11
12
X DIR.
-25.880
-70.710
-96.590
-96.590
-70.710
-25.880
25.880
70.710
96.590
96.590
70.710
25.880
RESULTS
?
l
l
l
l
l
l
l
l
l
l
l
l
BOUNDARY NODES
Y DIR.
-96.590
-70.710
-25.880
25.880
70.710
96.590
96.590
70.710
25.880
-25.880
-70.710
-96.590
?
1
1
1
1
1
1
1
1
1
1
1
1
X
-50
-86
-100
-86
-50
0
50
86
100
86
50
0
PRESCRIBED VALUES
DIR. ? Y DIR.
.000
.600
.000
.600
.000
.000
.000
.600
.000
.600
.000
.000
1
1
1
1
1
0
1
1
1
1
1
0
-86.600
-50.000
0.000
50.000
86.600
100.000
86.600
50.000
0.000
-50.000
-86.600
-100.000
?
1
1
1
1
1
1
1
1
0
1
1
1
X
-70
-96
-96
-70
-25
25
70
96
96
70
25
-25
DIR. ?
.710
.590
.590
.710
.880
.880
.710
.590
.590
.710
.880
.880
1
1
1
1
1
1
1
1
1
1
1
1
A A A i
Y DIR.
-70.
-25.
25.
70.
96.
96.
70.
25.
-25.
-70.
-96.
-96.
710
880
880
710
590
590
710
880
880
710
590
590
?
1
1
1
1
1
1
1
1
1
1
1
1
-0.7698500E+00
-0.1487200E+01
-O.21032OOE+O1
-0.2575850E+01
-0.2872950E+01
-0.2974500E+01
-0.2872950E+01
-0.2575850E+01
-0.2103200E+01
-0.1487200E+01
-0.7698500E+00
O.OOOOOOOE+00
0.7698500E+00
0.1487200E+01
0.21O320OE+01
0.2575850E+01
0.2872950E+01
-0.2872950E+01
-0.2575750E+01
-0.2103200E+01
-0.1487200E+01
-0.7698500E+00
O.OOOOOOOE+00
0.7698500E+00
0.1487200E+01
0.2103200E+01
0.2575850E+01
0.2872950E+01
0.2974300E+01
0.2872950E+01
0.2575750E+01
0.2103200E+01
0.1487200E+01
0.7698500E+00
DISPLACEMENT X
-0.4072179E-03
-0.7866786E-03
-0.1112431E-02
-0.1362535E-02
-0.1519704E-02
-0.1573262E-02
-0.1519712E-02
-0.1362557E-02
-0.1112482E-02
-0.7866530E-03
-0.4071935E-03
O.OOOOOOOE+00
0.4072177E-03
0.7866795E-03
0.1112433E-02
0.1362536E-02
0.1519705E-02
DISPLACEMENT Y
-0.1519695E-02
-0.1362574E-02
-0.1112502E-02
-0.7866730E-03
-0.4072844E-03
0.1928129E-08
0.4072882E-03
0.7866755E-03
0.1112481E-02
0.1362540E-02
0.1519707E-02
0.1573358E-02
0.1519692E-02
0.1362572E-02
0.1112499E-02
0.7866717E-03
0.4072846E-03
258
Chapter 4 Two Dimensional Elastostatics
0.2974500E+01
0.2872950E+01
0.2575850E+01
0.2103200E+01
0.1487200E+01
0.7698500E+00
O.OOOOOOOE+00
-0
-0
O.OOOOOOOE+00
0.7698500E+00
1487200E+01
0.2103200E+01
0.2575850E+01
2872950E+01
-0.2974300E+01
0.1573262E-02
0.1519712E-02
0.1362557E-02
0.1112481E-02
0.7866526E-03
0.4071933E-03
O.OOOOOOOE+00
O.OOOOOOOE+00
-0.4072880E-03
-0.7866766E-03
-0.1112482E-02
-0.1362543E-02
-0.1519709E-02
-0.1573361E-02
-0.76985E+00
-0.14872E+01
.21032E+01
.25759E+01
.28730E+01
.29745E+01
.2873OE+01
-0.2S759E+01
-0.21032E+01
-0.14872E+01
-0.76985E+00
.OOOOOE+00
.76985E+00
.14872E+01
.21032E+01
.25759E+01
.28730E+01
.29745E+01
.28730E+01
0.25759E+01
0.21032E+01
0.14872E+01
0.76985E+00
-0
0.OOOOOE+00 0.28730E+01
0.25758E+01
0.21032E+01
14872E+01
0.76985E+00
0.OOOOOE+00
0.76985E+00
0.14872E+01
0.21032E+01
0.25759E+01
0.28730E+01
0.29743E+01
0.28730E+01
0.25758E+01
O.21O32E+O1
0.14872E+01
0.76985E+00
0.OOOOOE+00
0.76985E+00
0.14872E+01
0.21032E+01
0.25759E+01
0.28730E+01
0.29743E+01
TRACTION X
BEFORE NODE
-0.25880E+02
-0.50000E+02
-0.70710E+02
-0.86600E+02
-0.96590E+02
-0.10000E+03
-0.96590E+02
86600E+02
0.70710E+02
0.50000E+02
0.25880E+02
0.32454E-03
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
0.96590E+02
0.86600E+02
0.70710E+02
0.50000E+02
0.25880E+02
0.24495E-03
-0
TRACTION Y
BEFORE NODE
-0.96590E+02
-0.86600E+02
-0.70710E+02
-0.50000E+02
-0.25880E+02
0.OOOOOE+00
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
0.96590E+02
0.86600E+02
0.70710E+02
0.50000E+02
0.25880E+02
0.61401E-03
-0.25880E+02
-0.50000E+02
-0.7071OE+02
-0.86600E+02
-0.96590E+02
-0.10000E+03
-0
TRACTION X
AFTER NODE
-0.25880E+02
-0.50000E+02
-0.70710E+02
-0.86600E+02
-0.96590E+02
-0.10000E+03
-0.96590E+02
-0.86600E+02
-0.70710E+02
-0.50000E+02
-0.25880E+02
-0.32454E-03
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
0.96590E+02
0.86600E+02
0.70710E+02
0.50000E+02
0.25880E+02
0.24495E-03 TRACTION Y
AFTER NODE
0.96590E+02
0.86600E+02
70710E+02
0.50000E+02
0.25880E+02
0.OOOOOE+00
0.25880E+02
0.50000E+02
0.70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
0.96590E+02
86600E+02
70710E+02
50000E+02
25880E+02
61401E-03
0.25880E+02
O.5O00OE+02
70710E+02
0.86600E+02
0.96590E+02
0.10000E+03
-0
INTERNAL POINTS DISPLACEMENTS
0.4000000E+01
0.2828430E+01
-0.4000000E+01
0.6000000E+01
0.100O000E+02
O.OOOOOOOE+00
0.2828430E+01
O.OOOOOOOE+00
O.OOOOOOOE+00
O.OOOOOOOE+00
DISPLACEMENT X
0.1169808E-02
0.8271520E-03
-0.1169808E-02
0.7798681E-03
0.4679200E-03
DISPLACEMENT Y
-0.4138201E-10
0.8271745E-03
-0.1176431E-08
-0.2637535E-10
-0.9115411E-09
INTERNAL POINTS STRESSES
0.4000000E+OX
0.2828430E+01
-0.4000000E+01
0.6000000E+01
0.1000000E+02
O.OOOOOOOE+00
0.2828430E+01
O.OOOOOOOE+00
O.OOOOOOOE+00
O.OOOOOOOE+00
SIGMA X
-0.5527559E+02
0.2465338E-02
-0.5527559E+02
-0.2456595E+02
TAU XY
-0.2949312E-03
-0.5527016E+02
-0.2649724E-03
-0.1923647E-03
-0.8843707E+01 -0.5847029E-04
SIGMA Y
0.5527643E+
0.3696308E0.5527644E+
0.2456613E+
0.8843761E+
**********************************
************************************
Example 4.3
The following example demonstrates the use of quadratic elements for a simple
problem which because of flexure, can not be solved accurately using constant
elements.
Figure 4.9 shows a rectangular plate under a linear distribution of tractions
in the horizontal direction, which represents two applied moments. The plate is
0.4000000E+01
0.4000000E+01
0.2000000E+01
O.OOOOOOOE+00
-0.2000000E+01
-0.4000000E+01
-0.4000000E+01
-0.2000000E+01
-0.2000000E+01
-0.2000000E+01
-0.2000000E+01
-0.2000000E+01
O.OOOOOOOE+00
0.2000000E+01
0.2000000E+01
O.2O0OO0OE+01
0.2000000E+01
0.2000000E+01
O.OOOOOOOE+00
BOUNDARY CONDITIONS
ELE
1
2
3
4
5
FIRST NODE
X DIR. ? Y DIR. ?
0.000
0.000
1000.000
0.000
0.000
6 1000.000
0.000
0.000
0.000
0.000
0.000
0.000
PRESCRIBED VALUES
SECOND NODE
X DIR. ? Y DIR. ?
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000
0.000
0.000
0.000
0.000
0.000
THIRD NODE
X DIR. ? Y DIR.
0.000
0.000
1-1000.000
1 0.000
1 0.000
1-1000.000
0.000
0.000
0.000
0.000
0.000
0.000
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
261
RESULTS
BOUNDARY NODES
-0.4000000E+01
-0.2000000E+01
O.OO000O0E+O0
0.2000000E+OX
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.2000000E+01
0.O0OOO00E+O0
-0.2000000E+01
-0.4000000E+01
-0.4000000E+01
-0
0.2000000E+01
0.2000000E+01
2000000E+01
0.2000000E+01
0.2000000E+01
O.OOOOOOOE+00
0.2000000E+01
0.2000000E+01
0.2000000E+01
0.2000000E+01
0.2000000E+01
O.OOOOOOOE+00
DISPLACEMENT X
-0.1999997E-01
-0.9999990E-02
O.OOOOOOOE+00
0.1000000E-01
0.2000001E-01
0.1396984E-07
-0.1999997E-01
-0.9999976E-02
O.OOOOOOOE+00
0.9999998E-02
0.2000002E-01
-0.2517481E-08
-0
0.40000E+01
0.20000E+01
0.O00OOE+OO
0.20000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.20000E+01
O.OOOOOE+00
20000E+01
0.40000E+01
0.40000E+01
-0.20000E+01
-0.20000E+01
-0.20000E+01
-0.20000E+01
-0.20000E+01
O.OOOOOE+00
0.20000E+01
0.20000E+01
0.20000E+01
0.20000E+01
0.20000E+01
0.00000E+00
INTERNAL POINTS DISPLACEMENTS
O.OOOOOOOE+00 O.OOOOOOOE+00
DISPLACEMENT Y
0.1999998E-01
0.5000000E-02
-0.2648449E-08
0.4999989E-02
0.1999998E-01
0.1874997E-01
0.1999997E-01
0.4999987E-02
0.000O000E+00
0.4999997E-02
0.2000000E-01
0.1875000E-01
TRACTION X
BEFORE NODE
-0.10000E+04
O.OOOOOE+00
-0.10431E-02
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
-0.10000E+04
O.OOOOOE+00
-0.18626E-03
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
TRACTION Y
BEFORE NODE
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
I8626E-03
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
-0
TRACTION X
AFTER NODE
O.OOOOOE+00
O.OOOOOE+00
10431E-02
O.OOOOOE+00
0.10000E+04
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
18626E-03
O.OOOOOE+00
0.10000E+04
O.OOOOOE+00
-0
-0
TRACTION Y
AFTER NODE
OOOOOE+00
00000E+00
OOOOOE+00
OOOOOE+00
OOOOOE+00
OOOOOE+OC
OOOOOE+Of
??????+??
18626E-03
OOOOOE+00
OOOOOE+00
O.OOOOOE+00
DISPLACEMENT X DISPLACEMENT Y
0.1111766E-07 -0.1250002E-02
INTERNAL POINTS STRESSES
X
Y
SIGMA X TAU XY SIGMA Y
O.OOOOOOOE+00 O.OOOOOOOE+00 0.1640320E-03 ?.6055832?-04 0.6484985E*****************************************************************************
Results for Uj and u2 displacements at corner A (figure 4.9(a)) are summarized
in Table 4.3, where they are compared against the analytical solution and the
values obtained using the program ELCONBE with 50 constant elements. It can
be seen that while the accuracy of the quadratic elements is excellent, poor res
ults
are obtained using constant elements. This demonstrates that constant elements
are not to be recommended for problems with flexural strains, although they give
good results in other applications.
Table 4.3 Displacements at Corner A (figure 4.9)
Displacement
2
Analytical
solution
-0.02
0.02
6 Quadratic element
solution
Values
-0.02
0.02
Error (%)
0%
0%
50 Constant
elements
Values Error (%)
-0.0157 21.5%
+ 0.0165 17.5%
(All results in m.)
262
Chapter 4 Two Dimensional Elastostatics
Example 4.4
Another flexural problem which can be solved accurately using quadratic elements
is the cantilever beam shown in figure 4.10. The beam has been discretized using
12 quadratic elements and is under a transverse parabolic load distribution at i
ts
two ends. On the end on the left hand side one also considers horizontal displac
edisplacement constraints as shown in the figure, with the lower corner node disp
lacements
13
???=1.5???
Figure 4.10 Cantilever beam under end loading
set both equal to zero. The problem is in plane stress and the material constant
s
are assumed to be the same as those in Example 4.3.
The beam theory gives for this problem a free end deflection of,
PI3 tn
u2 =
= 10 mm
3EJ
(a)
where P is the total load applied at the end, / is the length of the beam, E its
modulus of elasticity and / the moments of inertia of the beam.
The stress distributions along the fixed part of the boundary is linear in
accordance with the beam theory and equal to
(b)
21
Boundary Element Values of vertical displacements at nodes 11-12-13 at one end
and normal tractions at nodes 23-24-1 at the other end are given in Table 4.4. T
he
results agree well with those of formulae (a) and (b).
Table 4.4 Vertical Displacements and Tractions for the
Cantilever Beam of figure 4.10
Node
11
12
13
23
24
1
Vertical displacement
10.166
10.169
10.166
Normal tractions
On
29.85
-0.00
-29.85
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
0.O0OOOO0E+O0
0.2000000E+01
0.4000000E+01
0.6000000E+01
0.8000000E+01
0.1000000E+02
0.1200000E+02
O.14OOOOOE+O2
0.1600000E+02
0.1800000E+02
0.2000O0OE+O2
O.2OOOOOOE+O2
0.2000000E+02
0.1800000E+02
0.1600000E+02
0.14OOO0OE+O2
0.120OOOOE+02
0.1000000E+02
0.8000000E+01
0.6000000E+01
0.4000000E+01
0.2000000E+01
O.O0O0OO0E+OO
0.O00OO00E+O0
0.0000000E+00
0.0000000E+00
0.OOOO0O0E+0O
0.0000000E+0O
0.0000000E+00
0.0000000E+00
O.OOOOOOOE+00
0.0000000E+0O
0.O000000E+0O
O.OOOOOOOE+00
O.OOOOOOOE+00
0.2000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.400O0O0E+01
0.4000000?+??
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.2000000E+01
264
Chapter 4 Two Dimensional Elastostatics
BOUNDARY CONDITIONS
PRESCRIBED VALUES
?
LE
1
2
3
4
5
6
7
8
9
10
11
12
FIRST
X DIR. ?
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1
1
1
1
1
1
1
1
1
1
1
0
NODE
Y DIR.
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
?
0
1
1
1
1
1
1
1
1
1
1
1
SECOND NODE
X DIR. ? Y DIR.
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1
1
1
1
1
0.000
0.000
0.000
0.000
0.000
1 1500.000
1
1
1
1
1
0.000
0.000
0.000
0.000
0.000
0-1500.000
THIRD NODE
DIR. ? Y DIR.
**************************************************
1
0.000 1
1
0.000 1
1
0.000 1
1
0.000 1
1
0.000 1
1
0.000 1
1
0.000
1
0.000
1
0.000
1
0.000
1
0.000
1
0.000
***************
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
0.000 1
************
RESULTS
BOUNDARY NODES
X
0.0O00000E+00
0.2000000E+01
0.4000000E+01
0.6000000E+01
0.8000000E+01
0.1000000E+02
0.1200000E+02
0.14000O0E+02
0.1600000E+02
0.1800000E+02
0.2000000E+02
0.2000000E+02
0.2000000E+02
0.1800000E+02
0.1600000E+02
0.1400000E+02
0.1200000E+02
0.1000000E+02
0.8000000E+01
0.6000000E+01
0.4000000E+01
0.2000000E+01
0.0000000E+00
O.OO00000E+O0
0.0000000E+00
0.0000000E+00
0.0OOO00OE+OO
0.00OO00OE+0O
0.000000OE+0O
O.OOOOOOOE+00
0.0000000E+00
O.OOOOOOOE+00
O.00000O0E+00
O.OOOOOOOE+00
O.OOOOOOOE+00
0.2000000E+01
0.4000000E+01
1
1
1
1
1
0
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.4000000E+01
0.2000000E+01
DISPLACEMENT X
O.OOOOOOOE+00
0.2831624E+00
0.5382484E+00
0.7601910E+00
0.9555768E+00
0.1118159E+01
0.1254107E+01
0.1357250E+01
0.1433837E+01
0.1477282E+01
0.1492648E+01
0.1201406E-05
-0.1492655E+01
-0.1477283E+01
-0.1433836E+01
-0.1357249E+01
-0.1254107E+01
-0.1118160E+01
-0.9555784E+00
-0.7601921E+00
-0.5382496E+00
-0.2831629E+00
O.OOOOOOOE+00
O.OOOOOOOE+00
DISPLACEMENT Y
O.OOOOOOOE+00
0.1641325E+00
0.5995840E+00
0.1273359E+01
0.2155845E+01
0.3217330E+01
0.4428082E+01
0.5758359E+01
0.7178458E+01
0.8658765E+01
0.1016628E+02
0.1016907E+02
0.1016628E+02
0.8658760E+01
0.7178457E+01
0.5758362E+01
0.4428082E+01
0.3217330E+01
0.2155848E+01
0.1273359E+01
0.5995839E+00
0.1641333E+00
-0.2929009E-06
-0.4010320E-01
0.00000E+00
0.20000E+01
0.40000E+01
0.60000E+01
0.80000E+01
0.10000E+02
0.12000E+02
0.14000E+02
0.16000E+02
0.18000E+02
0.20000E+02
0.20000E+02
0.2000OE+O2
0.18000E+02
0.16000E+02
0.14000E+02
O.12OOOE+O2
0.10000E+02
O.OOOOOE+00
0.00000E+00
0.00000E+00
0.00000E+00
O.OOOOOE+00
0.00000E+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
0.20000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.40000E+01
TRACTION X
BEFORE NODE
-0.29853E+05
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
TRACTION Y
BEFORE NODE
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
0.15000E+04
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
TRACTION X
AFTER NODE
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
TRACTION Y
AFTER NODE
0.39985E-01
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+OC
O.OOOOOE+Of
O.OOOOOE+C!.
O.OOOOOE+00
0.15000E+04
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
O.OOOOOE+00
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
265
0.
0.
0.
0.
0.
0.
80000E+01
60000E+01
40000E+01
20000E+01
OOOOOE+00
00000E+0O
0.40000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.40000E+01
0.20000E+01
0.00000E+00
0.00000E+0O
0.00000E+00
0.OOOOOE+00
0.00000E+00
0.44191E-01
0.OOOOOE+00
0.00000E+00
0.00000E+00
0.OOOOOE+00
0.OOOOOE+00
-0.15000E+04
0.OOOOOE+00
0. OOOOOE+00
0.OOOOOE+00
0.00000E+00
0.29853E+05
0.44191E-01
0.OOOOOE+00
0.OOOOOE+00
0.OOOOOE+00
0. OOOOOE+00
0.OOOOOE+00
-0.15000E+04
INTERNAL POINTS DISPLACEMENTS
0.1000000E+02
Y DISPLACEMENT X
0.2000000E+01 -0.1154840E-06
DISPLACEMENT Y
0.3198673E+01
INTERNAL POINTS STRESSES
X
Y
SIGMA X TAU XY SIGMA Y
0.1000000E+02 0.2000000E+01 0.1745605E-01 0.1438650E+04 -0.1143646E*****************************************************************************
Example 4.5
This problem represents the case of a plane-strain hollow cylinder under interna
l
pressure as shown in figure 4.11. The pressure is assumed to be p = 100 N/mm2,
while the internal and external radii are rt = 10 mm and r2 = 25 mm, respectivel
y.
The elastic constants of the material are; Elasticity Modulus E = 200,000 N/mm2
and Poisson's ratio, v = 0.25.
Due to the symmetry of the problem only one quarter of the section needs to
be discretized. The boundary conditions are shown in figure 4.10(b). Three diffe
rent
discretizations consisting of 4, 10 and 15 quadratic elements are used as shown
in
figures 4.1 l(c).
Table 4.5 shows the relevant displacements computed for the points ?, ? and
? along the radius. The displacements are given in microns, i.e. 10" 3 mm and ar
e
compared with the exact values given by the theory of elasticity. It is interest
ing
to notice that even for the coarse discretization, results with 2% of the exact
solution are obtained. When 15 elements are employed, the results are within a
range of 0.1% of the exact solution.
Table 4.S Radial Displacements for Hollow Cylinder under Internal Pressure
(in 10 mm)
Node
A
?
?
Exact
value
8.0325
5.2912
4.4526
4 elements
7.8781
5.1668
4.3896
Discretization
10 elements
8.0246
5.2845
4.4520
15 elements
8.0350
5.2928
4.4631
266
Chapter 4 Two Dimensional Elastostatics
(a) Problem Definition
(i) 4 elements mesh
(c) Different quadratic meshes
(b) Boundary Conditions (including symmetry)
(ii) 10 elements mesh
Figure 4.11 Hollow cylinder under internal pressure
(iii) 15 elements mesh
Example 4.6
This example studies the stress concentration around a circular hole in a plate
stretching specimen as shown in figure 4.12(a). The geometric and material
constants are as follows:
G = 80,000 MPa
v = 0.25
(a)
Figure 4.12(b) shows the boundary discretization for half the plate using only
quadratic elements. Notice that only one element has been assumed over the
circular section.
4.8. Elastostatics Code using Quadratic Elements (ELQUABE)
267
1 t I 1 t t t
-* X,
??????
I(a) Square plate with hole. Geometry and boundary conditions
U2=0
n,=0
<r,=0
(b) Boundary element discretization
Figure 4.12 Square plate with hole. Geometrical definition and boundary
discretization
In order to check the accuracy of the Boundary Element solution the stress
concentration factor around the hole is computed. This factor can be defined as,
(b)
where ? is the applied stress at the top and bottom end of the plate.
268
Chapter 4 Two Dimensional Elastostatics
The boundary element solution gave a value of <?22 = 303.66 N/mm2 at node
A, which gives a stress concentration factor of ? = 3.0366. An experimental
solution for this problem gave ? = 3.03, which shows a good agreement with the
Boundary Element solution.
Exercises
4.1. Compute analytically the term G", that relates the first coordinate of a no
de with
itself in a linear elements program.
4.2.
Do the necessary transformations to the equation of exercise 4.1 to deco
mpose it into
integrals which may be computed either by standard Gaussian quadrature or by
logarithmic Gaussian quadrature.
4.3.
Discuss the effect of shifting the mid-node of a quadratic element from
the centre on
the numerical integrations (equation D.56)).
4.4.
Is there a limit of the distance from the mid-node to one end of a strai
ght line quadratic
element? Explain the reasons.
4.5.
The figure shows a link plate of a roller chain. Assuming that both pins
transmit their
force as a uniform pressure over one quarter of the hole, determine the normal s
tress er,,
along the central section of the link using 32 constant elements to model one qu
arter
of the problem (program ELCONBE).
p=25.4 mm
d = 7.9 mm
hi =20.8 mm
h2 = 13.8 mm
E=2105Nw/mm2
v=0.3
Plane stress
4.6.
Solve problem 4.5 using program ELQUABE and the same number of nodes with 16
quadratic elements. Solve the same problem using 32 quadratic elements. Compare
the
three solutions obtained for the problem. (The DIMENSIONS in the main program
ELQUABE must be changed to run the 32 elements example.)
Exercises
269
4.7. The link plate of a silent chain is shown in the figure. Using constant ele
ments and
the discretization of the figure compute the normal stresses along the central s
ection.
d=8 mm
p=25.4 mm
E=2.10s N/mm2
v=0.3
Plane stress
4.8.
Introduce the necessary changes in program ELQUABE to solve multiboundar
y
problems.
4.9.
Solve problem 4.7 with the same number of nodes but using quadratic elem
ents. Also
solve the problem with quadratic elements and double the number of nodes. Compar
e
the two solutions and also that of exercise 4.7.
4.10. The figure shows a centre cracked plate under traction. Because of the s
ymmetry only
one quarter of the plate is discretized. The a22 stress near the crack tip for p
oints
along the line A-B is known to be of the form a22 = KJ^Jlnr, r being the distanc
e
to the tip and K, the mode-/ stress intensity factor. Use the discretization A9
270
Chapter 4 Two Dimensional Elastostatics
quadratic elements) shown in the figure to solve the problem and represent the v
alue
of ozl^]2wr along the line A-B.
Comment on the results obtained near the tip. Compute the value of K, by
extrapolation to r = 0 of the last few nodes. Compare the value with the known
solution Ki = 6.67, exact within 1%.
1.8a
L/3 = 4a
Plane Stress
v=0.2
E=3.6.1O5 N/mm2
a = 10 mm
H
+ 0.4a 0.4a -|1.2a
Centre cracked plate under traction and discretization
of one quarter of the plate.
Chapter 5
Other Interesting Topics
5.1 Introduction
This chapter comprises a series of topics which are of general interest and exte
nd
the range of applications of boundary elements but are not as essential as those
described in previous chapters to understand the method.
One of the most interesting possibilities of boundary elements is that it is eas
y
to combine the technique with other numerical methods. The range of combinations
varies from those needed in analysis where boundary layers are ^interfaced to
boundary element regions, to simple coupling in static analysis. Sometimes for
instance, matrices for potential fluids are formed using boundary elements and
combined with finite element models for shells which represent a container, an
aerospace structure, an offshore platform, etc. The coupling is particularly sim
ple
as boundary elements accepts discontinuity of variables and full compatibility i
s
not required to obtain accurate answers. Ways of combining finite and boundary
elements are particularly attractive in view of the widespread use of both metho
ds.
Although many papers have been published on the topic, section 5.2 only discusse
s
comparatively simple ways of carrying out this combination. The first technique
is a purely intuitive approach which is not justified mathematically. It consist
s of
using the finite element results from a global solution as boundary conditions
when focusing on a particular region of the system. The finite element boundary
variables used are potentials or displacements as they are given with a higher
order of accuracy than fluxes or tractions.
The first of the other two approaches consists of treating the boundary element
region as a finite element and appropriately transforming the matrices. The seco
nd
approach consists of treating the finite element region as a boundary element an
d
manipulating the FE matrices in such a way that they can be implemented in the
boundary element system. Both approaches give similar results and using one or
the other will depend on which part of the problem (i.e. finite or boundary elem
ent
parts) is predominant.
Section 5.3 discusses special types of boundary element, which are produced
by asymptotic considerations when the boundary is far from the part under
perturbation. Under certain hypotheses these elements are equivalent to the
radiation type conditions presented by different authors. The section gives a
methodology of how these conditions can be obtained from basic boundary integral
considerations.
One of the first applications of boundary elements was the study of elastic
fracture mechanics problems for which singularities arise at the tip of the crac
k.
272
Chapter 5 Other Interesting Topics
These problems have been solved by different authors using a variety of boundary
integral formulations but more recently they have been studied using a simple
transformation which produces a singularity at the tip of the crack. This is ach
ieved
by using the so-called quarter point elements which were first developed for fin
ite
elements. They give a very elegant formulation in the case of boundary elements,
which directly produce the stress intensity coefficients in a way that can not b
e
done using finite elements.
The last section in this chapter explains how the technique can be expanded
to study steady state elastodynamics problems. Although this is similar to what
has been shown in Chapter 2 for the Helmholtz equation (section 2.14) it was
decided to include this amongst the special topics as the frequency dependent
formulation in elasticity is rather complex.
In spite of that, the implementation of the resulting relationships in existing
elastostatics codes (including those presented in Chapter 4 of this book) can be
attempted by the reader as explained in section 5.5.
5.2 Combination of Boundary and Finite Elements
There are sometimes advantages in combining finite and boundary element
solutions. In many unbounded field problems for instance, boundary elements
may provide the appropriate conditions to represent the infinite domain while
finite elements can solve complex material properties in the near domain. Bounda
ry
elements are also of interest in regions of high stresses or potentials, but fin
ite
elements may be adequate for other parts of the boundary and may be simpler
to use in cases such as layered continuum, anisotropic and non-linear materials.
Hence it is important for the analyst to be able to represent a body using finit
e
or boundary element techniques, depending on the particular geometries or
boundary conditions.
There are many papers written on the combination of the two techniques, but
from the viewpoint of simplicity of application as it relates to the codes alrea
dy
described here, we will consider only three methods, i.e.
(i) Method (i) Using the finite element solution to define the boundary conconditions for a localized boundary element region.
(ii) Method (ii) Treating a boundary element region as a finite element and
combining with finite elements.
(iii) Method (Hi) The converse of method (ii), i.e. treating a finite element re
gion
as an equivalent boundary element and combining with the other boundary
element region.
Method (i) This approach is a purely empirical technique and consists of having
solved a problem using finite elements to 'zoom' in a particular region using as
boundary conditions the finite element results for displacements or potentials.
The approach can not easily be justified from a mathematical standpoint but
it is used in several codes and seems to produce reasonable results. One can app
ly
5.2. Combination of Boundary and Finite Elements
273
this technique in the case of studying a region with a crack as shown in figure
5.1. First a global finite element solution is found using the mesh described in
figure 5.1 (a) and then the boundary element method is used to study the crack
region in more detail as shown in figure 5.1(b) using as boundary conditions the
displacements obtained in the finite element code. The reason why the approach
works is due to the fact that the finite element results for displacements (or
potentials) are usually accurate. The method would not give good results if the
finite element stresses (or fluxes) were used instead.
Method (ii) The second approach consists in treating the boundary element
region as a finite element. Consider the two regions as shown in figure 5.2 wher
e
region Q1 s expressed in terms of boundary solutions and fi2 discretized into fi
nite
elements.
The boundary element matrices for Q1 can be written as
HU = GP
E.1)
Figure 5.1. Method (i) combining global finite element solution with a localized
boundary element region
274
Chapter 5 Other Interesting Topics
while for region ?l2 the finite element matrices are
KU = F E.2)
where ? is the stiffness matrix of the problem and F the equivalent nodal forces
.
Note that U represents the displacements (or potentials) and P are the surface
tractions (or fluxes).
In order to combine E.1) and E.2) one can reduce the first to a finite element
form by inverting G, i.e.
G1HU = P
E.3)
Next one can convert the values of tractions at the nodes (as given by P) into a
n
equivalent nodal force matrix of the type used in finite elements. This is done
by
weighting the boundary tractions by the interpolation function used for the
displacements and produces a matrix M such that
F = MP E.4)
This operation is standard in finite elements although it is unusual to write th
e
distribution matrix M in an explicit form.
Equation E.3) can now be written as
M(G~1H)U = MP = F'
E.5)
where the right hand side vector has the same form as in finite elements.
One can now write formula E.5) as
K'U = F'
E.6)
where K^
K' is a stiffness matrix obtained from the boundary element formulation. It is
generally asymmetric due to the approximations involved in the discretization
process and the choice of the assumed solution. Although this matrix is sometime
s
symmetrized simply taking an average of the off-diagonal terms (i.e. assuming it
can be written as |(K' + K'T)) this is not recommended as it produces inaccurate
results in many practical applications. Obtaining symmetric boundary element
stiffness matrices may involve double integration of the type used in Galerkin's
BE formulation which are beyond the scope of this book.
The equivalent finite element type matrices of equation E.6) can now be
assembled with the matrices corresponding to region ?2 in figure 5.2 to form the
global stiffness matrix.
Method (Hi) This approach was proposed by Brebbia and Georgiou in 1979 [1]
and consists in treating the finite element region as an equivalent boundary ele
ment.
5.2. Combination of Boundary and Finite Elements
275
Consider the two regions described in figure 5.2. For region 1 one can write
the governing equations in a manner similar to that previously shown for
multiregion problems, i.e.
E.7)
where the subscript / defines the interface.
The matrices for the finite element region 2 can be written in a similar manner
using the concept of distribution matrix defined in formula E,4), i.e.
E.8)
By writing P, = P/1 = -P2 and U/ = U/1=U? one automatically satisfies the
equilibrium and compatibility conditions on the interface and equation E.7) and
E.8) can be rearranged and written together as follows.
??1 Hi -G,1 OH
L 0 K2 M,2 K2J
(?1)
?,
?2
E.9)
These equations will of course need to be rearranged in accordance with the
boundary conditions. Notice that this approach does not require any matrix
inversion.
Method (ii) - without forced symmetrization - and method (iii) are equivalent
and give the same numerical results. Using one or the other depends mainly upon
the problem in the sense of which part is more dominant, the finite elements or
the boundary elements, in which case one can use method (ii) or (iii) respective
ly.
Figure 5.2 Boundary divided into a finite elements and a boundary element region
276
Chapter 5 Other Interesting Topics
Method (ii) is essentially a stiffness method and can easily be incorporated in
existing finite element packages, although it does require the inversion of the
nonbanded G matrix. In contrast, method (iii) does not require this inversion and
both displacements and tractions remain unknown along the interface.
5.3 Approximate Boundary Elements
Combination of boundary elements with finite elements is particularly useful whe
n
dealing with domains tending to infinity. In these cases the near region is disc
retized
into finite elements and the outside domain is simulated with the boundary eleme
nts
on the interface of the two regions. This avoids having to discretize a larger r
egion
and at a certain distance putting some boundary conditions which try to represen
t
the domain going to infinity. The main drawback of this approach however, is
that the boundary element matrices are fully populated representing coupling of
all the nodes on the boundary. In practice this coupling can be avoided by assum
ing
that far from the region being perturbed the solution behaves in a smooth manner
.
This produces an approximate boundary element formulation that in many cases
is equivalent to the use of radiation or similar boundary conditions. It is impo
rtant
to point out that these approximate boundary elements are not related to the
so-called infinite elements which are based on domain rather than boundary
integration.
Consider the example shown in figure 5.3 where the internal region is assumed
to be subdivided into finite elements and the external region extending to infin
ity
is modelled using boundary elements. Let us consider that the problem is governe
d
by the Laplace's equation and hence the fundamental solution for two dimensions
is
E.10)
For any point in the internal domain (including those near but not on the ?7
interface) one can write,
| u*q dr = | uq* dT
E.11)
?;
?,
Substituting the fundamental solution E.10) into E.11) leads to
Notice that the reference point is considered to be outside the external region
and
hence c, = 0. The integration still needs to be carried out over all the ?; inte
rfaces
and all ? and q values are interrelated. One can simplify the formulation howeve
r,
if ?; is considered to be a circle of sufficiently large radius R which is assum
ed to
be constant, hence dT = R d9, where ? is the angular coordinate. Notice also tha
t
5.3. Approximate Boundary Elements
277
\
\
\
external region
(boundary elements)
\
Figure 5.3 Finite and boundary element regions
n = r if the reference point is far from the boundary ?; and hence equation E.12
)
can be written as
E.13)
q = du/dr.
This is a special form of the Sommerfeld condition which can also be written as,
1
du
- ? = (In r)
on ?,
r
dr
E.14)
Applying similar considerations for three dimensional potential problems one fin
ds
another form of this condition, i.e.
du 1
1- - ? = 0 oni,
dr r
E.15)
278
Chapter 5 Other Interesting Topics
The above radiation condition can then be applied as a boundary condition for
the finite element model describing the internal region.
The interesting aspect of this procedure is that it can be generalized to find
approximate radiation conditions in cases where the fundamental solution is
complex. Brebbia and Walker [2] have shown how given the Helmholtzequation
V2u + k2m = 0 E.16)
where ? is the wave number and ? the potential, the corresponding approximate
boundary elements (or radiation condition) on ?, are
+ ??? = 0 on ?,
E.17)
??
The demonstration follows the same steps as the one above.
Applications for wave diffraction problems demonstrate that the results
obtained using this approach are accurate. Other applications include harbour
resonance problems also governed by the Helmholtz equation [2] and soil
dynamics problems.
5.4 Singular Elements for Fracture Mechanics
Several approaches have been proposed in finite and boundary elements to model
the singular behaviour of stresses at a crack tip which occurs in elastic materi
als.
It is well known that the stresses near a traction free crack in an in-plane loa
ded
plate can be written as, [3]
*/ ?
0 . 3 ? ?? . e(. ? ??
cos -1 1
sin - sin - 0 I
sin -1 2 + cos - cos - 0 )
Bnr) 2\ 2 2 / VOr) 2\ 2 2/
df. . ? . 3 \ Ku 0.0 3.
- cos - 1 + sin - sin - 0 H
cos - sin - cos - 0 E.18)
) 2 V 2 2/ ?B??) 222
1?
? ? "\ tC
? /
? ^
ai2
..-' sm ? cos r cos r 0 + ..." cos - ( 1
sin -sin r
sin cos cos 0 Hcos I 1 sin sin
?B??) 2 2 2 ?B??) 2\
2 2.
where r and 0 are defined in figure 5.4, Kt and K,, are the stress intensity fac
tors
corresponding to the opening and sliding mode, respectively, and the size of r i
s
much smaller than the crack length. The displacements near the crack tip are,
Ki If ? ?(, ?
ie\ ?? \(' ?\ . 0/ . ,0\
Ml= - / Jcos- l-2v + sin2-) + /
sin- 2-2v + cos2li V\2w/ 2V
V it \l\2nj 2\ 2)
2?
1 + sin
2)
E.19)
5.4. Singular Elements for Fracture Mechanics 279
where ? is the shear modulus and v the Poisson ratio as in plane strain problems
.
Formulae E.18) and E.19) describe the stress and displacement distribution near
the crack tip and have been obtained analytically. Values of K, and Kn are diffi
cult
to obtain for general cases and it is then important to be able to model the beh
aviour
of cracks in boundary element codes.
Snyder and Cruse [4], Stern et al. [5] and Cruse [6] presented several
procedures to compute stress intensity factors using boundary elements, in
particular a singular quarter-point boundary element was proposed by Blandford
et al. [7] and by Martinez and Dominguez [8]. While some of the approaches
proposed are complex to implement, the quarter point element is easy to use in
boundary elements, gives accurate results and is no way sensitive to the discret
ization
used. As with other boundary element techniques, the domain needs to be divided
into subdomains by means of interfaces (figure 5.5) starting at the crack tip in
order to avoid having two similar sets of equations which will produce a singula
r
matrix. This subdivision avoids the numerical problems derived from having two
displacement variables for the same geometrical point along the crack. AH
boundaries are discretized into elements and elements are also defined along the
two faces of the crack and the interfaces between different regions. Boundary
conditions are applied at external boundaries including zero traction conditions
along the crack and the usual equilibrium and compatibility requirements are
satisfied at the interfaces.
The quarter point element is based on the quadratic expansion. In this case
any displacement, traction or coordinate such as x, and x2 can be represented as
f=0f
E.20)
where / represents geometrical displacements or traction variables as seen in
equations D.51) to D.54) and ? are the quadratic shape function matrices. Any
of these components can be written as,
? = ??+??? + ???
E-21)
For the particular case that the quadratic element has a straight-line geometry
and the mid-node is placed at a quarter of the length (figure 5.6) a simple
relationship can be found between the coordinate ?, and the variable f along the
element. In this case equation E.21) gives
fi = a}+af^-i + afri
E.22)
where
- 3/7 E.23)
Equation E.22) ensures that for this position of the mid-point, the yff behaviou
r
280
Chapter 5 Other Interesting Topics
2
2
/?
?
X
4
?'
? >
Ku1s
?,
1 Displacement
| correlation formulae
Nodal traction formula
***
0.2
0.8
1.0
04
06
//a
Figure 5.8 K, stress intensity factor computed using quarter-point singular
boundary elements
The value given by Bowie and Neal [8] is taken as reference as this value is
accurate within one percent. The results obtained by the quarter-point boundary
element procedure are given by Kr Results computed using singular quarter-point
elements and two different displacement correlation formulae (Kuls and Ku2s)
are given also for comparison. As can be seen in the figure, the nodal traction
procedure with singular quarter-point elements (Kt) gives accurate results for a
of them designed for special problems. Gaussian integration formulae are general
,
simple and very accurate and these formulae have been used for potential and
elasticity programs in Chapters 2 and 4.
In the following, Gaussian integral formulae for non-singular functions are
presented first. They can be used for integration over elements and internal cel
ls.
Formulae for one, two and three-dimensional domains are given. A second group
refers to integrals in which a logarithmic singularity is located at one end of
the
integration domain. This formula has been used for elements that include the
source point in the two-dimensional potential and elasticity programs of Chapter
s
2 and 4. In these cases the fundamental solutions present a logarithmic singular
ity
and the special integration formula is used for elements or cells in which the s
ource
is located.
Numerical integration formulae for functions with 1/r singularity and finite
part integrals for singular functions may be seen in references [1] and [2].
A.2 One-dimensional Gaussian Quadrature [3]
The integrals in this case can be written as,
Iiwlf(Zt) + EH (Al.l)
-1
i=l
where n is the number of integration points, ?{ is the coordinate of the ith int
egration
point, w( is the associated weighting factor and En is the error or residual, i.
e.
Formulae (A 1.1) is based on the representation of /(?) by means of Legendre
polynomials Pn(?). The ?f value is the coordinate at a point i where ? is zero an
d
for which the weights are given by
288
Appendix A Numerical Integration
Values of f j and w, are listed in Table Al.l. Notice that ?,- values are symmet
ric
with respect to i = 0, w.being the same for two symmetric values.
Table Al.l
1
0.577350269189626
i
0.00000 00000 00000
0.77459 66692 41483
i
0.33998 10435 84856
O.861I3 63115 94O53
J
0.00000 00000 00000
0.53846 9310105683
O.9O6I7 98459 38664
n = 6
0.23861 91860 83197
0.66120 93864 66265
0.9324695142 03152
t
0.00000 00000 00000
O.4O584 51513 77397
0.74153 11855 99394
0.94910 79123 42759
?
, = 2
1/5
1/5
0.333 333 33
0.797 426 99
0.101286 51
0.101286 51
O.O59 715 87
0.470 142 06
0.470 142 06
i
1/3
1/2
1/2
0
1/3
1/5
3/5
1/5
0.333 333 33
0.101 286 51
0.797 426 99
0.101286 51
0.470 142 06
0.059 715 87
0.470 142 06
ft
1/3
0
1/2
1/2
1/3
1/5
1/5
3/5
0.333 333 33
0.101 286 51
0.101286 51
0.797 426 99
0.470 142 06
0.470 142 06
O.O59 715 87
2w,
1
1/3
1/3
1/3
-9/16
25/48
25/48
25/48
0.225 000 00
0.125 939 18
0.125 939 18
0.125 939 18
0.132 394 15
0.132 394 15
0.132 394 15
@.0.1)
Figure A 1.1
A.0.0)
Triangular coordinates
290
Appendix A Numerical Integration
@.0.1.1)
@.1.0.1)
@.0.1.-1)
A,0.0.-1)
Figure A 1.2 Pentahedral cell
Integration over a pentahedral cell (figure A1.2) can be done by combination
of (A 1.5) and the one-dimensional Gaussian quadrature (A 1.1), i.e.
/=
t=u
(A 1.6)
where n is the number of integration points over the base and m the number of
integration points along the height; ?[, ?\, ?'3 and w\l) are given in Table A1.
2
and ?{ and wJ-2) are shown in Table A 1.1.
A.5 One-dimensional Logarithmic Gaussian Quadrature [3]
Kernels including a logarithmic singularity at one end of the integration domain
can be integrated using the following formula [3].
(A 1.7)
where integration point coordinates ?, and weighting factors wt are given in Tab
le
A1.3.
Appendix A Numerical Integration
Table A 1.3
291
n
2
3
4
5
6
7
ii
0.11200880
0.60227691
0.63890792 (-1)
0.36899706
0.76688030
0.41448480 (-1)
0.24527491
0.55616545
0.84898239
0.29134472 (-1)
0.17397721
0.41170251
0.67731417
0.89477136
0.216344005 (-1)
0.12958339
0.31402045
0.53865721
0.75691533
0.92266884
0.16719355 (-1)
0.10018568
0.24629424
0.43346349
0.63235098
0.81111862
0.94084816
0.71853931
0.28146068
0.51340455
0.39198004
0.94615406
0.38346406
0.38687532
0.19043513
0.39225487
0.29789346
0.34977622
0.23448829
0.98930460
0.18911552
0.23876366
0.30828657
0.24531742
0.14200875
0.55454622
0.10168958
0.19616938
0.27030264
0.23968187
0.16577577
0.88943226
0.33194304
0.59327869
n
8
9
10
ft
0.13320243
0.79750427
0.19787102
0.35415398
0.52945857
0.70181452
0.84937932
0.95332645
0.10869338
0.64983682
0.16222943
0.29374996
0.44663195
0.60548172
0.75411017
0.87726585
0.96225056
0.90425944
0.53971054
0.13531134
0.24705169
0.38021171
0.52379159
0.66577472
0.79419019
(-1)
(-1)
(-1)
(-1)
(-1)
(-1)
(-1)
(-1)
(-2)
(-1)
(-1)
(-1)
(-1)
(-2)
(-1)
0.89816102
0.96884798
w,0.16441660
0.23752560
0.22684198
0.17575408
0.11292402
0.57872212
0.20979074
0.36864071
0.14006846
0.20977224
0.21142716
0.17715622
0.12779920
0.78478879
0.39022490
0.13867290
0.24080402
0.12095474
0.18636310
0.19566066
0.17357723
0.13569597
0.93647084
0.55787938
0.27159893
0.95151992
0.16381586
(-1)
(-D
(-2)
(-1)
(-1)
(-1)
(-2)
(-1)
(-1)
(-D
(-2)
(-2)
Note: Numbers are to be multiplied by the power of 10 in parentheses.
A.6 Variable Transformation Methods
To deal with singular and nearly singular integrals, there are methods using tra
nstransformation of variables in order to weaken or cancel out the singularity by
the
Jacobian of the transformation before applying the ordinary Gauss rule. This
approach has been proposed by Telles [5] using quadratic and cubic transformatio
n
and by Hayami and Brebbia [6] who developed the idea of transforming in the
radial direction working in polar coordinates.
References
[1] Brebbia, ? A., Telles, J. ? F. and Wrobel,,L. ? Boundary Element Techniques,
Springer-Verlag, Berlin, N.Y., 1984.
[2] Kutt, H. R. Quadrature Formulae for Finite Part Integrals, Report WISK 178,
The
National Research Institute for Mathematical Sciences, Pretoria, 1975.
tener'a
Plate Model, in Boundary Elements in Engineering |C. A, Brebbia, Ed.}, SpringerVcriag, Btriin, 19S2.
[172] Weeen, Van ??? F. Application of the Boundary Integral Equation Method to
Reissner* Plate Model. Iru- Jnl Numer, Meth ????., 1982,19, 1-10.
[173] Long, S. Y,, Brebbia, ? A. and Telles, J.CF. Boundary Element Bending Anal
ysis
of Moderately Thick Plates, Engng. Analysis, 1967, *C).
[174] Kiuniya, N.. Sawpki. Y, and NTikamura, J. Boundary Element Nonlinear Bendi
ng
Analysis of Clamped Sandwich Plates and Shdls, in Boundary Element* in Engineeri
ng
(C. A. Brebbia. Ed). Springer-Verlag, Berbn and CML Publications, Southampton,
?02
Appendix C tcferatces for Further Raiding
[175] ??????, ?. Integral Equation Approach to Small and Large Displacements of
Thin
Elastic Plates, in Boundary Elements in Engineering (? A. Brebbia, Ed). SpnngerVerlag, Berlin and CML Publications, Southampton. 1932,
[176] Kim, J. W. On the Computation of 111* Stress Intensity Factors in Elastic
Plate
Fltiure via Boundary Integra] Equations, in Boundary Elements in Engineering (C.
A,
Brebbia. Ed. |, Springer-Verlag, Berlin and CM L Publications, Southampton, 1982
,
[177] TanaLs, M. and Miyazaki, ? A Direct BEM for Elastic-Plate Structures subje
cted
to Arbitrary Loadings in Boundary Elements Vil (?. ?. Brebbia and G, Maier,
Eds.}, Springer-Verlag, Berlin and CML Publications, Southampton, 1985
[17ft] Costa. J A. and Brebbia, C. A. Elastic Buckling of Plales using the Bound
ary
Element Method, in Boundary Elements VII [C A. Brebbia and G. Maier, Eds.).
Springer-Verlag, Berlin and CML Publications, Southampton, 1985[179] Antes, H. On Boundary Integral Equations for Circular Cylindrical Shells,
in
Boundary Element Methods (C. A. Brebbia, Ed.}, Springer-Vertag, Berlin and NY,
19SL
[ISO] Karnita, N-, Sawaki, Y. and Nakamura I. Finite and PostbuckbiVB. Deformati
on of
Heated Plates and Shallow Shells, in Boundary Elements (C, A. Brebbia et a/., Ed
s),
Springer-Vcrlag, Berlin and CML Publications, Southampton. 1?[181] Tosak, N. and Miy eke. S- Non-linear Analysis ?? Elastic Shallow Shells by
Boundary
Element Methods, in Boundary Element Methods Vtl {?, A. Brebbi* and G. Maier,
Eds.), Springer-Verlag, Berlin and CML Publications, Southampton, 1985.
[182] McDonald, ?. ?. and Wexler, A. Finite Element Solutions of Unbounded Field
Problems, IEEE Trans. Micrewaue Theory MTT-20, 1972, 41-847.
[I S3] Shaw, R P. and Falby, *, FTBIE: A Combined Finite Element - Boundary Inte
gral
Equation Method, Ptoc- )si Symposium on Innovative Numerical Analysb f Applied
Engineering Science. Versailles, CETIM, 1977.
[184] Osias, J. W , Wilson, R. B. and Scitclraan, L. A. Combined Boundary Integr
al
Equation Finite Element Analysis of Solids, Prdc. 1st Symposium on innovative
Numerical Analysis in Applied Engineering Science. Versailles. ????, 19T9,
[?] Brebbia. ? A. and Georgiou, P. Combination of Boundary and Finite Elements f
or
Field
by using Boundary Element Method. BEM Viti (?, ?. ?????? ft at., Eds. I, CM
Publications. Southampton and Springer-Vetlag, Berlin and NY, 1936
?217] Brcbbia. C. A. and Magurcanu. R. The Boundary Element Method for Ekctmmagnetic Problems, Engineering Analysis, Jrd.t I9B7,4D),
Enokuono. M , Kagawa. R. and Nakaroura T. Non Linear Analysis of Magnetic
Field by Boundary Element Method taking imo Account Extern*! Power Sources.
BEM VU (? .?. ?????? et a/, Eds), CM Publications, Southampton and SprtngerVerlag, Berlin, 1986.
Wendland. W, Asymptotic Accuracy and Convergence, in Progress in Boundary
Eknttnt Methods, Voi. I (C.A.Bitbbta, Ed. I, Pentech Press. London.] 981,289-312
,
Wendland, W, On Asymptotic Error Analysis and Underlying Mathematical
Principles for Boundary Element Methods, in Boundary Element Techniques in
Computer Aided Engineering (C. A. Brebbii. Ed.). NATO ASf Series E-84 Martimis
Nijhoff PubL. DordTUht/Boston/Uncaiter, 1984, 417-436.
Wendland. W. Asymptotic Accuracy and Convergence for Point Collocation
Methods, Topics in Boundary Element Research 2 |C- A. Brcbbia, Ed.}, Springer
Vcrlag, Berlin, 1985, 230-257.
[222] Wcndland, W. On Some Mathematical Aspects at Boundary Element Methods for
Elliptic Problems, in Tht Mathematics of Finite Elements and Applications V
{i. Whileman, Ed), Academic Press. London. 1985, 193^227.
Wendland. W.. Hsiao, G ? and Kopp, P. The Synthesis of the Collocation and
the Galenn Method Applied lo Some Integra] Equations of tbe Firsi Kind, in Hew
Dtvftopmems in Boundary Element Methods (C- A. Brabbia, Ed.), CML Pubtkationt,
Sotithampton. 1980, 122-136.
[224] Wendlund. W. and Schwab, Ch. 3-D BEM and Numerical Integration, in Boundar
y
Ettmenis in Engineering VII {C. A. Bitbbta and G, Maier, Eds.J, Sphnficr-Verlag.
Berlin nd CML Publications. Southampton, 198$, Vol. |[. 13,85-13.102.
[2251 Wendland, W. On Asymptotic Error Estimate* lor the Combined Boundary and
Finite Element Method, in Innovative Numerical Methods in Engineering (R, P. Sha
w
et at., Eds.I, Springer-Vcrlag, Berlin and CML Publications, Southampton, I486.
55-70.
[226] Wendland, W, On the Asymptotic Convereence of Boundary Integral Methods, u
i
Boundary Element Methods (C. A. Brebbia, Ed.? Springer-Verlag. Berlin and CML
Publics I ions, Southampton, 1981,
[227] Wend land, W. ami Saranan, J. On the Asymptotic Conwergeiwe of Collocation
Methods with Spline Functions of Even D^grw1, of*. ????. 19*5,43, 91-108.
[228] Wendland, W. *nd Arnold, D. N. Collocation versus Galerkin Procedures for
Boundary Integral Methods, in Boundary Element Methods in Engineering {C. A.
Brebbia, Ed), Springcr-Verlag. Berlin and CML Publications, Southampton. 1982.
18-31.
[229] Wendland, W.. Hsiao, G, ? and Siephan . H. On the Integral Equation Method
for the Plane Mixed Boundary Value Problem with the Laplacian, MtKhennolKtu
Methods in the Applied Sciences, 1979, 1, 265-321.
[23OJWendland. W. and Stephan. E. Boundary Element Method for Membrancand Torsio
n
Crack Problems, Computer Meih. Appi. Mech- Eng,, 1983, 36. 331-35S,
[231] Moia Scares, ? A., Rodriguez. H. C, Olivejr Farias. L. M. and Hang. E. ?
Optimization of the Shape of Solids and Hollow Shafts using Boundary Elements.
in Boundary Elements (CA Brebbia et ai., Eds, ? Springer- Verla(, Berlin and CML
Publications, Southampton, 1983.
[232] Miyamoto. Y.. Iwaiaki. S. and Sugimom. H. On Study of Shape Optimization o
f
2-Dimensional Elastic Bodies by BEM. jti Boundary Elements ? III (M ?????? and
? A. Brebbia. Eds). Springer- Vtrlag, Berlin and CML Publications, Southampton,
[233] Kine, J. H. Shape Optimization Utilizing a Boundary Element Formulation, i
n
ii ? Reference for Further ReadJit|
?05
BETECH/&6 |C ?. ????? and JL J. Connor, Ed*,), Computational Mechanics
Publications, Southampton. (??.
[234] Novati, G. and Bnjbbi, ?, ?. Boundary Element Formulation for Geometrically
Nonlinear Elaitoitatk*, Appl. Math. Mod., 1982. *
123$] Brebb, C. A. and Nftu. 5. M. State of the Art: Computational Applications o
f
Boundary Element Metbodj for Calhodk Projection of Offshore Sliuctures. OMAE*
88 Conference Proceeding*. ASME, 19*8,
[236] Brebbia. ? A. and Trevelyan, J. On the Accuracy and Convergence of Boundar
y
Element Results for Lhe Floyd Pressure Veud Problem, Technical Note in Computers
and Structures. Vd. 24, No. 3. S13-SI6, Pergamon pre, 1986.
[237] Brebbia. C. A. Counted Systems, Chapter 5 in Fhute Element Handbook (H,
????????? ei ?< Eds), McGraw-Hill Book Co.. NY, 1987.
[238] Khatawa. K,, Wiobel, L. C. Brebbia, ? A. and Tanika, M. A Boundary Element
FormulniioD for Natural Convection Problems. Int. Jnl. for Numerical Methods in
Fluids; 3, I9S8.
Bfcbbia, ? ?- and Tang. W. On the Treatment of Boundary Force Integrals in
Boundary Elements, CMWR Conference, Morocco, Contpuiational Mechanics Pubiications, Southampton and Springer-Verlag, Berlin and NY. (??.
[240] Brebbia, ? A. and Umetam. M. Convergence and Accuracy Studies of Adaptive
BEM Solutions, in Proc. Relkibilfty and Robustness ????????, Come, Computational
Mechanics Publications, Southampton, 1987,
Brebbia, C. A. BEASY: A Boundary Element System at Structural Analysis,
Structural Analysis Systems, Vol. I (A. Niku-Lari, Ed.), Pergamoa Press, Onford,
1986.
[242] Brebbia, ? A. and DeFigueriredo. T. Critical Companion of Finite and Bound
ary
Element Methods, NAFEMS, Brighton international Cvnf. on Quality Assurance and
Standards, in Finite Elemnt Aneiysis, 1987.
[243] Brebbia, ? A. and Wrobcl, L. ? The Solution of Parabolic Problems using th
e
Dual Reciprocity Boundary Element Method, Proc. of WTAttt Symposium, San
Aatip. Texas, (9S7. Spnnjer-Veriag. Btrtin and NY, 1903.
[244] Wrobel. L. C. and Brebbia, ?. ?. The Dual Reciprocity Boundary Element
Formulation tor Nonlinear Diffusion Problems, ? output ? ? Methods Jn Applied
Mechanics a*d ?>m, ??, 65, 147-164.
Aksundri, ? and ??????, ?. A. Strength of Masonry Walls under Static Horizontal
Loads: Boundary Element Analysis and Experimental Tests, Jnt, Jnf, Eng- Analysis
,
), ?7.
Further refertuce to rtcent boundary dement work am be found in
the following books
[24*1 Brebbia, C. A. and Connor, J. J. (Eds) Advances in Boundary Elements C Vol
s)
Proceeding* of the 11th Inrt. Conference. Computttional Mechanics Publications.
Southampton, 198,
B47| Tanaka, M., Horuna, T. and Brebbia. C. A. (Ed*) Boundary Element Methods.
Proceedings of the 12th but. Conference. Computational Mechanics Publications.
Southampton, 1990.
[248] Brebbia, C. A, and GipsOii. 5, <Eds> Boundary Elements X11E B volumes) Pro
Proceedings of die 13th Intrn. Conference. Computaiional Mechanics Publications,
Soulhampton, 1991.
[249] Brebbi*. C, A. vtd Zarruuii, N. Boundary Element Techniques. Applications
m
Enjinering Proceedbige of ????????, ComputMkinat Mechanics Publicatioas.
Southampton, 1999.
[250) Grilli, S., Cheng, A. and Brebbia, C. A. Computational Engineering with Bo
unBoundary Element! B vols) Proceedings of BETECH/90, ???????????] Mechanics
Pubticmtioiu, SotHDemptort. t99O.
306
[2511 ??????, C. A. {Ed) Boundary Element Technology VI pTuceeduigt of
??????/1, Computational Mechanics Publications, Southampton. 1991.
[252) ?????, ? A. and Chmidouet-Mintwk, A. Boundary Element! in Mechanical and
Electric*! Engineering, Computational Mechanics PubUcaitofts, Southampton,
1989
[253] Bnbbii, ? A.. ??????, M. and Shi*, R, Advances in Boundary Elctneflts in
Japan and USA', Computational Mechanics Publications. Southampton, 1990,
[254 J Brebbia, ? A. (Ed), Topic in Boundary Element Heaearch, Vol. S. Viscous Ro
w
Application!, Springer-Verlftg, Berlin uid NY, 1989.
[255] Brtobia. ? A. (Ed), Topics in Boundary Element Reuarcn, Vol, 6, Ekctnnugnedc AppllcatKiut. Springer-Vedag, Beriin and KY. 1900.
[256] Brebbia, C. A. (Bd). Topics in Boundacy Element Retesrch. Vol. 7, Electiof
tatia
Appli?atiooi. Springer-Verlag, Berlin and NY, 1900
1257 ] Ciikowiki. R. D. end Birbbia, ? A. Advances ui DoundaTy Elenicot Methods
?
Acoustics, Comnutatiooal Mtdmracs Publicaikms, SotilhanqNon. 1991.
[258 ] Partridge, P, W.. Brebbia. ? A. and Wrabel, L. C. The Dual Reciprocity Bo
unBoundary Element Method, Compulationi) Mechanics PuMkatiora, Southampton,
1991.
[259] Aliabadi, M and Brebbin, ? A. Advances in Boundary Element Methods for
Fnctwe MechaDks, Computational Hechuucs Publkatiods, Southampton. 1992.
[260] Wrobtl, L. ? and Brebbia, ? A. Advances in Boundary Element Medud tor
Heat Tnuiifu, ?????????] Meduuks Publicitkms, Southampton, 1992.
[261} Alitbadt, M. and Rookie, D. Numerical Fncture Mechanics, Contputttioul Med
ufiki
Publicttioiu, Southampton and Boston. Kluwti Acadenic Publishers, Dordrecfai, 19
91.
Other Source of References
The International Society of Boundary Elements ASBE) publishes a bi-monthly
Newsletter and Abstracts which lists the work In boundary elements. For further
information of how to join the Society and obtain copy of (h* bulletin write to
1SBE
Asbursi Lodge
Ashurst
Southampton SO4 ?? A, UK
Fax: 0703 292853
Appendix D
Answers to Selected Exercises
Chafittr 1
1.1. u1JL u-x{ltA u-1.333*-?-???*1
1.7. b(!>
-7/32
?
??
Chapw 2
No, because the only difference between the equations ??? two different scales w
ould
be the term
1?. The soluiion \\ obtained locating (wo point iourctJ in the complete space; o
ik t the
coKocation point and the other at the mirror image ? the firai with ????? la the
free surface
where r' is the distance from the poinl iniage of the collocation point to the p
oint of
interest.
2,7. ? - In rr. The imigf source is now negative.
???
Appendix D Answer* to Selected Excrcis**
2?.
2??1
111. q
Sn
??
when:
VB
Pi
Pi
P*J
??? ??? Saw 7c- 16 16 16
l
??
16
??
cot ~
?
??*
2Sx
15*
16
3x
S*
7?
5oo 7??
16
16
iv.
7* 21n 33*
CM CCH ? CM
\6
L
t*
5 15* 25t 35
cos cos
cos
0
16 16 H5 16
3i 9? 1
i eoe cat16
16
??
16
?
??
5?
7?
i cos ??* ?
16
16
16
1?
Chapter i
Al. mJ is of the order of - when ? ^0 and p, is (Inilt. Since ihe area of the su
rface of
E
integration is of the order of eJ, the integral wili a I way shave a zero limit
whtot -^0.
Appendix D ??? wen to Selected Exencua
309
3J. For an edge along the x, axis defined by the scmT-plmeJ ??, ?3 >0 and xlh x^
,;
i 0
0
1
0
?? n* is symmetric, This Tact ??? be interpreted as a consequence of the recipro
city
relation. It cin also be verified by simple inspection of equations 13,51) and 0
-54).
p* is rat symmetric,
3.6. When a system of Cartesian coordinates with two axes parallel to the sides
of the
element i& adopted all the off diagonal terms of H" and G" are zero.
x1 j .
310
Appendix D Answers to Selected Exercises
Similarly,
J J P*iZtti\G\dfd4
-i -1
where f,. i;. art the coordinates given in table Al. I of Appendix A and w,wm
the corresponding weighting Tutors.
?? ? rigid body motion would, change the term of the basic equation
by adding a constant to uj. Thui ?? additional term of the form
$ consum pt dr
conuani j fi Jr
?
?
would be introduced.
Fur any bounded region in equilibrium,
The above Answer corresponds to problems with zero body forces, but the same can
easily be shown to be true for nonaro body forces.
3,9, When the tractions over the internal boundaries have ? ato rttultouL Thus,
f
where rh, arc the internal boundaries.
3.1 a
f
uj, cos ? - uJi sin ? uj, sin &
uj, coi 9
?]?
(if, M $ + ??? un fi - 7, sin 0 + jif t cm 0 k?i
ir|, ??? ? + uj, lin fl u|, iin fl + uj, coi 0 u^j
11 the ? are measured from points with $' = --.
Appendix D Anjwen to Selected
111.
j<>+
?.
?
??
?
where
s = sin ?
?? ? ?3 ?
i1 ~ iin3 ?
Fundamental solution with the collocation point locattd alff
?/2.
Chapter 4
4?.
- J
?|. j
i-i
312
???????, 129
for potential pnatokmj, 142
Higher Ofdtr. 148
Gulcrtun method, 23. 2ft. 29,
?
Calertin't vecur, 1*0, 163
force. 149
Hanktlt fuottknu, 134
Hrlmhpte eqiatiiM, IW
Higher Order ??????
indirect rprmulilioni in
(lailMiaiict, 193
Ltdirect fimrailniwn in
. 8
Initial Hmm or iifiim, 159
IKS
Integnli, evibilkm, 55
Intcrml cell 123. 129
Intenal poinci In <Uutcnuticin
IM
Irttmil pcirii in p
???????. 55
[menial pointt in Pm
217
120.
rruHiplicn, 17
HrttMion, 17, 46
Linev detneMi. 51, 70, ?
lor potenlii]
, 74
OrtJwtrop> end niHtropy, [21
2?
Plane unto. 209
Point cuHootiua mcihod, 33.
Poimn't cqoatiori. 17. IZ3
Pueaiiil ??????. 49
Quadntic dementi, 51, 19,
23J
Coupukn- ?*** for
daHortatm, 241
??????? ?? for ?????
??????, 44
Quadrilatenl ind iriaajuUr
donc4i fonbree
IBS, 124
Kifjd body coutUjoat, 17?
Single Layer
Suit (rf tuaia,
StrttK* 00 die ???????. Ill
Svbdoimia coUocuke, 23.15
SubrtfHHd ia elatutarka, IM
Suhregi?*in potentia)
H, 131
Three
dHooMinuhie* at
confer point, 183
Tradiau and ditpUcanem
in
, ISO
Trefftz method, 37
VtrUbtf uimfbrmtiDn
mrthcdi 291
Wctk tonnulpikw. 31
Weighted residual tecfarbquet,
22
Computational Mechanics Publications
Boundary Element Methods
for Damage Tolerance Design
of Aircraft Structures
N.K. SALGADO, EmbraerSA, Brazil.
This book presents 2D formulations for the
analysis of Aircraft stiffened panels and sheets
repaired with mechanically fastened or
adhesively bonded patches. The possibility
of using such formulations in Interactive
systems for Damage Tolerance Design is
investigated and an object orientated from
such systems is presented.
Part 1I Coottmtsi Damage Tolerance
Assessment; Boundary Element Method;
Analysis of Stiffened Panels; A Design
System of Stiffened Panels; Analysis of
Mechanically Attached Repairs and Lap
Joints; Analysis of Adhesively Bonded
Patches.
ISBN: 1853 L25326
199$
???
?69.00/$UO
A Green's Function TimeDomain BEM of
Elascodynunics
Edited by; ? WCHTER, University of Bochum,
Germany.
This CD demonstrates how the transient
Green's Function of the elastodynamic 2D
Lamb's problem is derived and used to develop
a fast and accurate time-domain BEM- The
Green's function is purely algebraic without
any integrals and is presented in numerically
applicable form for the first time.Partial
Contents: Boundary Element Method; TimeDomain BEM; Cagniard-de-Hoop Method:
Elastodynamics; Wave Propagation; Green's
FunctionISBN: 1?5312494?
1998
BooL an CD ?59.00/594.00
Boundary Element
Formulations for Thick
Places
Y.F. RASHED, Caifo University, Egypt.
Plate Bending Problems have many
applications in Civil and Mechanical
Engineering. The analysis of plates uamg the
BEM has not received much attention in
existing research, instead the classical plate
theory which ignores shear deformations has
been utilised. Et has been demonstrated dial
the modelling using the classical plate theory
shows many difficulties in dealing with curved
boundaries and problem with stress
concentrations. This work shows Boundary